{ "schema_version": "a3s.gateway.ai-comparison.v1", "commit": "fa360f6566759cc689b37dc602e8b87370dfec11", "run_url": "https://github.com/A3S-Lab/Gateway/actions/runs/34691275387", "generated_at": "2026-09-12T11:46:40Z", "environment": { "runner_image": "ubuntu24-20260907.300.1", "cpu_model": "AMD EPYC 7763 64-Core Processor", "logical_cpus": 4, "memory_mib": 15993, "kernel": "Linux 6.17.0-1022-azure x86_64 GNU/Linux" }, "versions": { "a3s_gateway": "a3s-gateway 1.1.1", "nginx": "nginx version: nginx/1.24.0 (Ubuntu)", "deterministic_upstream": "a3s-gateway 1.1.1", "streaming_load_client": "a3s-gateway 1.1.1" }, "methodology": { "scope": "Same-host OpenAI-compatible streaming comparison against one deterministic upstream. A3S performs bounded OpenAI JSON/model validation; NGINX is a transport-only baseline.", "trials": 5, "aggregation": "Median of repeated trials; A3S and NGINX execution order alternates by trial for every profile. Every raw trial remains embedded.", "clock": "Load-generator monotonic clock; stored latency unit is microseconds.", "token_definition": "One decoded non-terminal SSE data event with the next exact benchmark_sequence value; network chunks are never counted as tokens.", "correctness_gate": "100% HTTP success, SSE content type, exact monotonic sequence, exact token count, one [DONE] marker, and no post-terminal bytes.", "nginx_streaming": "HTTP/1.1 upstream keep-alive, response buffering/cache/compression disabled, request buffering enabled with a 512 KiB client-body buffer for the core prompt profiles, and 120-second timeouts.", "threshold": "Ratios within 3% are marked within threshold." }, "profiles": { "stream-overhead-c1": { "label": "Streaming overhead, C1", "workload": "Zero-delay 32-token chat stream at concurrency 1", "capability_alignment": "a3s_openai_validation_vs_nginx_transport", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 1, "requests": 128, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "products": { "a3s-gateway": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:46:46.486366177+00:00", "product": "a3s-gateway", "trial": 1, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 1, "requests": 128, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 5.257962722, "streams_per_second": 24.344029573361436, "token_goodput_per_second": 779.0089463475659, "ttft": { "samples": 128, "min_us": 336, "mean_us": 462.3125, "p50_us": 453, "p90_us": 533, "p95_us": 582, "p99_us": 670, "max_us": 676 }, "inter_token_latency": { "samples": 3968, "min_us": 1, "mean_us": 1307.7149697580646, "p50_us": 10, "p90_us": 38, "p95_us": 56, "p99_us": 40088, "max_us": 40934 }, "time_per_output_token": { "samples": 128, "min_us": 1295, "mean_us": 1307.7265625, "p50_us": 1306, "p90_us": 1312, "p95_us": 1329, "p99_us": 1331, "max_us": 1332 }, "end_to_end": { "samples": 128, "min_us": 40800, "mean_us": 41030.7734375, "p50_us": 40975, "p90_us": 41130, "p95_us": 41716, "p99_us": 41842, "max_us": 41844 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:31.415352509+00:00", "product": "a3s-gateway", "trial": 2, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 1, "requests": 128, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 5.25698742, "streams_per_second": 24.348545996710794, "token_goodput_per_second": 779.1534718947454, "ttft": { "samples": 128, "min_us": 353, "mean_us": 474.96875, "p50_us": 461, "p90_us": 582, "p95_us": 629, "p99_us": 655, "max_us": 685 }, "inter_token_latency": { "samples": 3968, "min_us": 1, "mean_us": 1307.251008064516, "p50_us": 9, "p90_us": 36, "p95_us": 53, "p99_us": 40112, "max_us": 40948 }, "time_per_output_token": { "samples": 128, "min_us": 1294, "mean_us": 1307.25, "p50_us": 1306, "p90_us": 1313, "p95_us": 1327, "p99_us": 1331, "max_us": 1332 }, "end_to_end": { "samples": 128, "min_us": 40780, "mean_us": 41025.171875, "p50_us": 40964, "p90_us": 41136, "p95_us": 41746, "p99_us": 41824, "max_us": 41901 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:04.792415972+00:00", "product": "a3s-gateway", "trial": 3, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 1, "requests": 128, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 5.257069303, "streams_per_second": 24.348166748906184, "token_goodput_per_second": 779.1413359649979, "ttft": { "samples": 128, "min_us": 357, "mean_us": 477.140625, "p50_us": 459, "p90_us": 578, "p95_us": 612, "p99_us": 686, "max_us": 828 }, "inter_token_latency": { "samples": 3968, "min_us": 1, "mean_us": 1307.1814516129032, "p50_us": 9, "p90_us": 38, "p95_us": 57, "p99_us": 40100, "max_us": 40844 }, "time_per_output_token": { "samples": 128, "min_us": 1296, "mean_us": 1307.203125, "p50_us": 1305, "p90_us": 1312, "p95_us": 1326, "p99_us": 1335, "max_us": 1373 }, "end_to_end": { "samples": 128, "min_us": 40805, "mean_us": 41028.7421875, "p50_us": 40968, "p90_us": 41126, "p95_us": 41694, "p99_us": 42002, "max_us": 43126 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:49.689386196+00:00", "product": "a3s-gateway", "trial": 4, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 1, "requests": 128, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 5.213897887, "streams_per_second": 24.549771164323534, "token_goodput_per_second": 785.5926772583531, "ttft": { "samples": 128, "min_us": 278, "mean_us": 468.03125, "p50_us": 457, "p90_us": 559, "p95_us": 569, "p99_us": 635, "max_us": 960 }, "inter_token_latency": { "samples": 3968, "min_us": 1, "mean_us": 1296.702620967742, "p50_us": 9, "p90_us": 37, "p95_us": 55, "p99_us": 40074, "max_us": 40959 }, "time_per_output_token": { "samples": 128, "min_us": 13, "mean_us": 1296.6953125, "p50_us": 1306, "p90_us": 1312, "p95_us": 1314, "p99_us": 1332, "max_us": 1332 }, "end_to_end": { "samples": 128, "min_us": 1369, "mean_us": 40691.4921875, "p50_us": 40970, "p90_us": 41078, "p95_us": 41183, "p99_us": 41818, "max_us": 41823 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:23.055409317+00:00", "product": "a3s-gateway", "trial": 5, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 1, "requests": 128, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 5.258021843, "streams_per_second": 24.343755850007792, "token_goodput_per_second": 779.0001872002493, "ttft": { "samples": 128, "min_us": 342, "mean_us": 461.4375, "p50_us": 450, "p90_us": 546, "p95_us": 590, "p99_us": 656, "max_us": 761 }, "inter_token_latency": { "samples": 3968, "min_us": 1, "mean_us": 1308.053679435484, "p50_us": 9, "p90_us": 37, "p95_us": 55, "p99_us": 40111, "max_us": 41092 }, "time_per_output_token": { "samples": 128, "min_us": 1294, "mean_us": 1308.046875, "p50_us": 1306, "p90_us": 1313, "p95_us": 1331, "p99_us": 1334, "max_us": 1337 }, "end_to_end": { "samples": 128, "min_us": 40742, "mean_us": 41037.6171875, "p50_us": 40973, "p90_us": 41150, "p95_us": 41753, "p99_us": 41865, "max_us": 41980 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 24.348166748906184, "token_goodput_per_second": 779.1413359649979, "ttft": { "samples_per_trial": 128, "min_us": 342, "mean_us": 468.03125, "p50_us": 457, "p90_us": 559, "p95_us": 590, "p99_us": 656, "max_us": 761 }, "inter_token_latency": { "samples_per_trial": 3968, "min_us": 1, "mean_us": 1307.251008064516, "p50_us": 9, "p90_us": 37, "p95_us": 55, "p99_us": 40100, "max_us": 40948 }, "time_per_output_token": { "samples_per_trial": 128, "min_us": 1294, "mean_us": 1307.25, "p50_us": 1306, "p90_us": 1312, "p95_us": 1327, "p99_us": 1332, "max_us": 1332 }, "end_to_end": { "samples_per_trial": 128, "min_us": 40780, "mean_us": 41028.7421875, "p50_us": 40970, "p90_us": 41130, "p95_us": 41716, "p99_us": 41842, "max_us": 41901 }, "cancellation": null, "fault": null } }, "nginx": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:46:52.239150750+00:00", "product": "nginx", "trial": 1, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 1, "requests": 128, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 5.207956192, "streams_per_second": 24.577779704948792, "token_goodput_per_second": 786.4889505583614, "ttft": { "samples": 128, "min_us": 327, "mean_us": 454.4296875, "p50_us": 437, "p90_us": 521, "p95_us": 544, "p99_us": 588, "max_us": 1046 }, "inter_token_latency": { "samples": 3968, "min_us": 1, "mean_us": 1296.5352822580646, "p50_us": 1, "p90_us": 3, "p95_us": 9, "p99_us": 40520, "max_us": 40692 }, "time_per_output_token": { "samples": 128, "min_us": 3, "mean_us": 1296.375, "p50_us": 1306, "p90_us": 1310, "p95_us": 1312, "p99_us": 1314, "max_us": 1314 }, "end_to_end": { "samples": 128, "min_us": 1163, "mean_us": 40658.046875, "p50_us": 40974, "p90_us": 41054, "p95_us": 41101, "p99_us": 41132, "max_us": 41137 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:25.611270417+00:00", "product": "nginx", "trial": 2, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 1, "requests": 128, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 5.249145053, "streams_per_second": 24.384923393733466, "token_goodput_per_second": 780.3175485994709, "ttft": { "samples": 128, "min_us": 384, "mean_us": 464.4375, "p50_us": 457, "p90_us": 524, "p95_us": 531, "p99_us": 591, "max_us": 591 }, "inter_token_latency": { "samples": 3968, "min_us": 1, "mean_us": 1306.4715221774193, "p50_us": 1, "p90_us": 2, "p95_us": 9, "p99_us": 40485, "max_us": 41370 }, "time_per_output_token": { "samples": 128, "min_us": 1299, "mean_us": 1306.3515625, "p50_us": 1306, "p90_us": 1310, "p95_us": 1310, "p99_us": 1311, "max_us": 1336 }, "end_to_end": { "samples": 128, "min_us": 40816, "mean_us": 40976.7421875, "p50_us": 40971, "p90_us": 41048, "p95_us": 41067, "p99_us": 41138, "max_us": 41874 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:10.585161811+00:00", "product": "nginx", "trial": 3, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 1, "requests": 128, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 5.247935376, "streams_per_second": 24.390544248195788, "token_goodput_per_second": 780.4974159422652, "ttft": { "samples": 128, "min_us": 346, "mean_us": 442.9765625, "p50_us": 435, "p90_us": 493, "p95_us": 530, "p99_us": 611, "max_us": 680 }, "inter_token_latency": { "samples": 3968, "min_us": 1, "mean_us": 1306.9032258064517, "p50_us": 1, "p90_us": 2, "p95_us": 8, "p99_us": 40518, "max_us": 40685 }, "time_per_output_token": { "samples": 128, "min_us": 1290, "mean_us": 1306.7265625, "p50_us": 1307, "p90_us": 1311, "p95_us": 1311, "p99_us": 1312, "max_us": 1314 }, "end_to_end": { "samples": 128, "min_us": 40697, "mean_us": 40967.5, "p50_us": 40963, "p90_us": 41048, "p95_us": 41062, "p99_us": 41168, "max_us": 41191 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:43.928282377+00:00", "product": "nginx", "trial": 4, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 1, "requests": 128, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 5.249178185, "streams_per_second": 24.3847694798724, "token_goodput_per_second": 780.3126233559168, "ttft": { "samples": 128, "min_us": 348, "mean_us": 497.5078125, "p50_us": 491, "p90_us": 544, "p95_us": 583, "p99_us": 614, "max_us": 1398 }, "inter_token_latency": { "samples": 3968, "min_us": 1, "mean_us": 1305.3424899193549, "p50_us": 1, "p90_us": 2, "p95_us": 8, "p99_us": 40467, "max_us": 41505 }, "time_per_output_token": { "samples": 128, "min_us": 1274, "mean_us": 1305.1796875, "p50_us": 1305, "p90_us": 1310, "p95_us": 1311, "p99_us": 1313, "max_us": 1340 }, "end_to_end": { "samples": 128, "min_us": 40806, "mean_us": 40974.1796875, "p50_us": 40966, "p90_us": 41057, "p95_us": 41079, "p99_us": 41119, "max_us": 41981 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:28.849236899+00:00", "product": "nginx", "trial": 5, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 1, "requests": 128, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 5.248049161, "streams_per_second": 24.390015427296415, "token_goodput_per_second": 780.4804936734853, "ttft": { "samples": 128, "min_us": 329, "mean_us": 459.21875, "p50_us": 455, "p90_us": 532, "p95_us": 538, "p99_us": 569, "max_us": 576 }, "inter_token_latency": { "samples": 3968, "min_us": 1, "mean_us": 1306.407258064516, "p50_us": 1, "p90_us": 2, "p95_us": 9, "p99_us": 40501, "max_us": 40724 }, "time_per_output_token": { "samples": 128, "min_us": 1300, "mean_us": 1306.25, "p50_us": 1306, "p90_us": 1310, "p95_us": 1311, "p99_us": 1314, "max_us": 1315 }, "end_to_end": { "samples": 128, "min_us": 40823, "mean_us": 40969.1875, "p50_us": 40965, "p90_us": 41044, "p95_us": 41074, "p99_us": 41110, "max_us": 41161 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 24.390015427296415, "token_goodput_per_second": 780.4804936734853, "ttft": { "samples_per_trial": 128, "min_us": 346, "mean_us": 459.21875, "p50_us": 455, "p90_us": 524, "p95_us": 538, "p99_us": 591, "max_us": 680 }, "inter_token_latency": { "samples_per_trial": 3968, "min_us": 1, "mean_us": 1306.407258064516, "p50_us": 1, "p90_us": 2, "p95_us": 9, "p99_us": 40501, "max_us": 40724 }, "time_per_output_token": { "samples_per_trial": 128, "min_us": 1290, "mean_us": 1306.25, "p50_us": 1306, "p90_us": 1310, "p95_us": 1311, "p99_us": 1313, "max_us": 1315 }, "end_to_end": { "samples_per_trial": 128, "min_us": 40806, "mean_us": 40969.1875, "p50_us": 40966, "p90_us": 41048, "p95_us": 41074, "p99_us": 41132, "max_us": 41191 }, "cancellation": null, "fault": null } } }, "comparison": { "a3s_to_nginx_token_goodput_ratio": 0.9982841881131655, "a3s_to_nginx_stream_rate_ratio": 0.9982841881131655, "a3s_to_nginx_ttft_p50_ratio": 1.0043956043956044, "a3s_to_nginx_ttft_p99_ratio": 1.1099830795262267, "a3s_to_nginx_inter_token_latency_p50_ratio": 9.0, "a3s_to_nginx_inter_token_latency_p99_ratio": 0.9900990099009901, "a3s_to_nginx_time_per_output_token_p50_ratio": 1.0, "a3s_to_nginx_time_per_output_token_p99_ratio": 1.0144706778370145, "a3s_to_nginx_end_to_end_p50_ratio": 1.0000976419469805, "a3s_to_nginx_end_to_end_p99_ratio": 1.0172614995623845, "positions": { "token_goodput": "within_threshold", "stream_rate": "within_threshold", "ttft_p50": "within_threshold", "ttft_p99": "nginx_lower", "inter_token_latency_p50": "nginx_lower", "inter_token_latency_p99": "within_threshold", "time_per_output_token_p50": "within_threshold", "time_per_output_token_p99": "within_threshold", "end_to_end_p50": "within_threshold", "end_to_end_p99": "within_threshold" } } }, "stream-overhead-c64": { "label": "Streaming overhead, C64", "workload": "Zero-delay 32-token chat streams at concurrency 64", "capability_alignment": "a3s_openai_validation_vs_nginx_transport", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 256, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "products": { "a3s-gateway": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:46:52.720385611+00:00", "product": "a3s-gateway", "trial": 1, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 256, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 256, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.166992349, "streams_per_second": 1533.0043653676612, "token_goodput_per_second": 49056.13969176516, "ttft": { "samples": 256, "min_us": 309, "mean_us": 4730.93359375, "p50_us": 4815, "p90_us": 7124, "p95_us": 7619, "p99_us": 8966, "max_us": 9396 }, "inter_token_latency": { "samples": 7936, "min_us": 1, "mean_us": 1058.5824092741937, "p50_us": 4, "p90_us": 263, "p95_us": 805, "p99_us": 39487, "max_us": 42993 }, "time_per_output_token": { "samples": 256, "min_us": 142, "mean_us": 1058.5859375, "p50_us": 1298, "p90_us": 1376, "p95_us": 1398, "p99_us": 1408, "max_us": 1419 }, "end_to_end": { "samples": 256, "min_us": 11000, "mean_us": 37629.37890625, "p50_us": 44004, "p90_us": 49382, "p95_us": 49806, "p99_us": 51144, "max_us": 51363 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 256 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:32.360489635+00:00", "product": "a3s-gateway", "trial": 2, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 256, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 256, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.150038849, "streams_per_second": 1706.2247658271492, "token_goodput_per_second": 54599.192506468775, "ttft": { "samples": 256, "min_us": 248, "mean_us": 3370.38671875, "p50_us": 2054, "p90_us": 6926, "p95_us": 11298, "p99_us": 12109, "max_us": 12279 }, "inter_token_latency": { "samples": 7936, "min_us": 1, "mean_us": 1023.2791078629032, "p50_us": 4, "p90_us": 214, "p95_us": 522, "p99_us": 40000, "max_us": 41641 }, "time_per_output_token": { "samples": 256, "min_us": 45, "mean_us": 1023.28125, "p50_us": 1302, "p90_us": 1347, "p95_us": 1355, "p99_us": 1418, "max_us": 1442 }, "end_to_end": { "samples": 256, "min_us": 2961, "mean_us": 35157.390625, "p50_us": 41944, "p90_us": 45102, "p95_us": 46483, "p99_us": 48872, "max_us": 49316 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 256 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:11.066068431+00:00", "product": "a3s-gateway", "trial": 3, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 256, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 256, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.169093287, "streams_per_second": 1513.957203990008, "token_goodput_per_second": 48446.63052768026, "ttft": { "samples": 256, "min_us": 234, "mean_us": 3626.015625, "p50_us": 2851, "p90_us": 8235, "p95_us": 9175, "p99_us": 13121, "max_us": 15203 }, "inter_token_latency": { "samples": 7936, "min_us": 1, "mean_us": 1049.272933467742, "p50_us": 4, "p90_us": 201, "p95_us": 428, "p99_us": 40089, "max_us": 43613 }, "time_per_output_token": { "samples": 256, "min_us": 23, "mean_us": 1049.28125, "p50_us": 1303, "p90_us": 1358, "p95_us": 1376, "p99_us": 1431, "max_us": 1439 }, "end_to_end": { "samples": 256, "min_us": 4998, "mean_us": 36210.4609375, "p50_us": 42463, "p90_us": 45987, "p95_us": 47065, "p99_us": 49462, "max_us": 52012 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 256 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:50.656567899+00:00", "product": "a3s-gateway", "trial": 4, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 256, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 256, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.166071162, "streams_per_second": 1541.5078507128167, "token_goodput_per_second": 49328.25122281013, "ttft": { "samples": 256, "min_us": 260, "mean_us": 3246.84765625, "p50_us": 2831, "p90_us": 6995, "p95_us": 7514, "p99_us": 8328, "max_us": 8944 }, "inter_token_latency": { "samples": 7936, "min_us": 1, "mean_us": 1068.823714717742, "p50_us": 4, "p90_us": 246, "p95_us": 693, "p99_us": 40091, "max_us": 42528 }, "time_per_output_token": { "samples": 256, "min_us": 123, "mean_us": 1068.859375, "p50_us": 1307, "p90_us": 1365, "p95_us": 1383, "p99_us": 1415, "max_us": 1427 }, "end_to_end": { "samples": 256, "min_us": 9947, "mean_us": 36504.7578125, "p50_us": 42350, "p90_us": 45770, "p95_us": 46542, "p99_us": 50639, "max_us": 52233 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 256 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:29.338563490+00:00", "product": "a3s-gateway", "trial": 5, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 256, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 256, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.18515533, "streams_per_second": 1382.6229036992886, "token_goodput_per_second": 44243.932918377235, "ttft": { "samples": 256, "min_us": 291, "mean_us": 2780.50390625, "p50_us": 1760, "p90_us": 6532, "p95_us": 8239, "p99_us": 10128, "max_us": 10373 }, "inter_token_latency": { "samples": 7936, "min_us": 1, "mean_us": 1271.461441532258, "p50_us": 4, "p90_us": 194, "p95_us": 381, "p99_us": 40511, "max_us": 42745 }, "time_per_output_token": { "samples": 256, "min_us": 6, "mean_us": 1271.46484375, "p50_us": 1333, "p90_us": 1371, "p95_us": 1388, "p99_us": 1440, "max_us": 1451 }, "end_to_end": { "samples": 256, "min_us": 2625, "mean_us": 42268.71484375, "p50_us": 42975, "p90_us": 48000, "p95_us": 50096, "p99_us": 50249, "max_us": 50688 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 256 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 1533.0043653676612, "token_goodput_per_second": 49056.13969176516, "ttft": { "samples_per_trial": 256, "min_us": 260, "mean_us": 3370.38671875, "p50_us": 2831, "p90_us": 6995, "p95_us": 8239, "p99_us": 10128, "max_us": 10373 }, "inter_token_latency": { "samples_per_trial": 7936, "min_us": 1, "mean_us": 1058.5824092741937, "p50_us": 4, "p90_us": 214, "p95_us": 522, "p99_us": 40089, "max_us": 42745 }, "time_per_output_token": { "samples_per_trial": 256, "min_us": 45, "mean_us": 1058.5859375, "p50_us": 1303, "p90_us": 1365, "p95_us": 1383, "p99_us": 1418, "max_us": 1439 }, "end_to_end": { "samples_per_trial": 256, "min_us": 4998, "mean_us": 36504.7578125, "p50_us": 42463, "p90_us": 45987, "p95_us": 47065, "p99_us": 50249, "max_us": 51363 }, "cancellation": null, "fault": null } }, "nginx": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:46:53.194448593+00:00", "product": "nginx", "trial": 1, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 256, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 256, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.165211191, "streams_per_second": 1549.5318352859038, "token_goodput_per_second": 49585.01872914892, "ttft": { "samples": 256, "min_us": 275, "mean_us": 3065.8671875, "p50_us": 2442, "p90_us": 6725, "p95_us": 7923, "p99_us": 9185, "max_us": 10112 }, "inter_token_latency": { "samples": 7936, "min_us": 1, "mean_us": 997.6708669354839, "p50_us": 1, "p90_us": 2, "p95_us": 5, "p99_us": 40476, "max_us": 42271 }, "time_per_output_token": { "samples": 256, "min_us": 2, "mean_us": 997.69921875, "p50_us": 1297, "p90_us": 1330, "p95_us": 1337, "p99_us": 1349, "max_us": 1365 }, "end_to_end": { "samples": 256, "min_us": 2347, "mean_us": 34030.5, "p50_us": 41770, "p90_us": 44333, "p95_us": 44783, "p99_us": 45341, "max_us": 46371 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 256 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:31.896225451+00:00", "product": "nginx", "trial": 2, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 256, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 256, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.171117552, "streams_per_second": 1496.0475825413864, "token_goodput_per_second": 47873.522641324365, "ttft": { "samples": 256, "min_us": 252, "mean_us": 2638.6640625, "p50_us": 1645, "p90_us": 6580, "p95_us": 7747, "p99_us": 9895, "max_us": 10359 }, "inter_token_latency": { "samples": 7936, "min_us": 1, "mean_us": 1074.801285282258, "p50_us": 1, "p90_us": 2, "p95_us": 5, "p99_us": 40526, "max_us": 42232 }, "time_per_output_token": { "samples": 256, "min_us": 2, "mean_us": 1074.86328125, "p50_us": 1296, "p90_us": 1339, "p95_us": 1345, "p99_us": 1361, "max_us": 1365 }, "end_to_end": { "samples": 256, "min_us": 1970, "mean_us": 35977.0, "p50_us": 41684, "p90_us": 44153, "p95_us": 45009, "p99_us": 46388, "max_us": 48780 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 256 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:11.549186594+00:00", "product": "nginx", "trial": 3, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 256, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 256, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.172015572, "streams_per_second": 1488.2373556273149, "token_goodput_per_second": 47623.595380074075, "ttft": { "samples": 256, "min_us": 301, "mean_us": 2353.19140625, "p50_us": 1122, "p90_us": 6127, "p95_us": 7572, "p99_us": 8363, "max_us": 8641 }, "inter_token_latency": { "samples": 7936, "min_us": 1, "mean_us": 1110.589339717742, "p50_us": 1, "p90_us": 2, "p95_us": 5, "p99_us": 40530, "max_us": 42389 }, "time_per_output_token": { "samples": 256, "min_us": 1, "mean_us": 1110.63671875, "p50_us": 1298, "p90_us": 1336, "p95_us": 1344, "p99_us": 1359, "max_us": 1368 }, "end_to_end": { "samples": 256, "min_us": 1650, "mean_us": 36798.24609375, "p50_us": 41436, "p90_us": 43542, "p95_us": 44238, "p99_us": 46444, "max_us": 46717 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 256 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:50.174277818+00:00", "product": "nginx", "trial": 4, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 256, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 256, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.176084592, "streams_per_second": 1453.8466829624706, "token_goodput_per_second": 46523.09385479906, "ttft": { "samples": 256, "min_us": 265, "mean_us": 2832.37109375, "p50_us": 1494, "p90_us": 7326, "p95_us": 8359, "p99_us": 9429, "max_us": 40458 }, "inter_token_latency": { "samples": 7936, "min_us": 1, "mean_us": 1056.780241935484, "p50_us": 1, "p90_us": 2, "p95_us": 5, "p99_us": 40560, "max_us": 42375 }, "time_per_output_token": { "samples": 256, "min_us": 2, "mean_us": 1056.78515625, "p50_us": 1300, "p90_us": 1338, "p95_us": 1348, "p99_us": 1357, "max_us": 1368 }, "end_to_end": { "samples": 256, "min_us": 1218, "mean_us": 35609.375, "p50_us": 41176, "p90_us": 43984, "p95_us": 44568, "p99_us": 45387, "max_us": 80739 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 256 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:29.815457116+00:00", "product": "nginx", "trial": 5, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 256, "token_count": 32, "first_token_delay_ms": 0, "token_interval_ms": 0, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 256, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.169263475, "streams_per_second": 1512.4349774811135, "token_goodput_per_second": 48397.91927939563, "ttft": { "samples": 256, "min_us": 330, "mean_us": 2501.765625, "p50_us": 1329, "p90_us": 6674, "p95_us": 7864, "p99_us": 9483, "max_us": 9604 }, "inter_token_latency": { "samples": 7936, "min_us": 1, "mean_us": 1043.3399697580646, "p50_us": 1, "p90_us": 2, "p95_us": 5, "p99_us": 40650, "max_us": 42008 }, "time_per_output_token": { "samples": 256, "min_us": 2, "mean_us": 1043.37109375, "p50_us": 1299, "p90_us": 1329, "p95_us": 1339, "p99_us": 1356, "max_us": 1366 }, "end_to_end": { "samples": 256, "min_us": 716, "mean_us": 34888.72265625, "p50_us": 41456, "p90_us": 43266, "p95_us": 43866, "p99_us": 44636, "max_us": 45269 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 256 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 1496.0475825413864, "token_goodput_per_second": 47873.522641324365, "ttft": { "samples_per_trial": 256, "min_us": 275, "mean_us": 2638.6640625, "p50_us": 1494, "p90_us": 6674, "p95_us": 7864, "p99_us": 9429, "max_us": 10112 }, "inter_token_latency": { "samples_per_trial": 7936, "min_us": 1, "mean_us": 1056.780241935484, "p50_us": 1, "p90_us": 2, "p95_us": 5, "p99_us": 40530, "max_us": 42271 }, "time_per_output_token": { "samples_per_trial": 256, "min_us": 2, "mean_us": 1056.78515625, "p50_us": 1298, "p90_us": 1336, "p95_us": 1344, "p99_us": 1357, "max_us": 1366 }, "end_to_end": { "samples_per_trial": 256, "min_us": 1650, "mean_us": 35609.375, "p50_us": 41456, "p90_us": 43984, "p95_us": 44568, "p99_us": 45387, "max_us": 46717 }, "cancellation": null, "fault": null } } }, "comparison": { "a3s_to_nginx_token_goodput_ratio": 1.0247029461212023, "a3s_to_nginx_stream_rate_ratio": 1.0247029461212023, "a3s_to_nginx_ttft_p50_ratio": 1.894912985274431, "a3s_to_nginx_ttft_p99_ratio": 1.0741329939548203, "a3s_to_nginx_inter_token_latency_p50_ratio": 4.0, "a3s_to_nginx_inter_token_latency_p99_ratio": 0.989119170984456, "a3s_to_nginx_time_per_output_token_p50_ratio": 1.0038520801232667, "a3s_to_nginx_time_per_output_token_p99_ratio": 1.044952100221076, "a3s_to_nginx_end_to_end_p50_ratio": 1.024290814357391, "a3s_to_nginx_end_to_end_p99_ratio": 1.1071231850529888, "positions": { "token_goodput": "within_threshold", "stream_rate": "within_threshold", "ttft_p50": "nginx_lower", "ttft_p99": "nginx_lower", "inter_token_latency_p50": "nginx_lower", "inter_token_latency_p99": "within_threshold", "time_per_output_token_p50": "within_threshold", "time_per_output_token_p99": "nginx_lower", "end_to_end_p50": "within_threshold", "end_to_end_p99": "nginx_lower" } } }, "stream-paced-c16": { "label": "Paced chat stream, C16", "workload": "50 ms first token and 10 ms cadence at concurrency 16", "capability_alignment": "a3s_openai_validation_vs_nginx_transport", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "products": { "a3s-gateway": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:46:55.448000977+00:00", "product": "a3s-gateway", "trial": 1, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.592040686, "streams_per_second": 40.19997765308367, "token_goodput_per_second": 1286.3992848986775, "ttft": { "samples": 64, "min_us": 50994, "mean_us": 51927.984375, "p50_us": 51970, "p90_us": 52459, "p95_us": 52523, "p99_us": 52532, "max_us": 52532 }, "inter_token_latency": { "samples": 1984, "min_us": 9766, "mean_us": 11138.76814516129, "p50_us": 11381, "p90_us": 11609, "p95_us": 11667, "p99_us": 11777, "max_us": 12071 }, "time_per_output_token": { "samples": 64, "min_us": 11005, "mean_us": 11138.75, "p50_us": 11132, "p90_us": 11207, "p95_us": 11227, "p99_us": 11231, "max_us": 11231 }, "end_to_end": { "samples": 64, "min_us": 392792, "mean_us": 397254.625, "p50_us": 397198, "p90_us": 399814, "p95_us": 400470, "p99_us": 400609, "max_us": 400609 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:36.867668937+00:00", "product": "a3s-gateway", "trial": 2, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.594233851, "streams_per_second": 40.1446751114056, "token_goodput_per_second": 1284.629603564979, "ttft": { "samples": 64, "min_us": 50797, "mean_us": 51825.5, "p50_us": 51853, "p90_us": 52219, "p95_us": 52322, "p99_us": 52410, "max_us": 52410 }, "inter_token_latency": { "samples": 1984, "min_us": 10040, "mean_us": 11157.313508064517, "p50_us": 11365, "p90_us": 11600, "p95_us": 11655, "p99_us": 11836, "max_us": 13131 }, "time_per_output_token": { "samples": 64, "min_us": 11012, "mean_us": 11157.3125, "p50_us": 11167, "p90_us": 11253, "p95_us": 11254, "p99_us": 11257, "max_us": 11257 }, "end_to_end": { "samples": 64, "min_us": 392569, "mean_us": 397733.03125, "p50_us": 398175, "p90_us": 400645, "p95_us": 400747, "p99_us": 400943, "max_us": 400943 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:13.797029788+00:00", "product": "a3s-gateway", "trial": 3, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.591902149, "streams_per_second": 40.20347609945968, "token_goodput_per_second": 1286.5112351827097, "ttft": { "samples": 64, "min_us": 51093, "mean_us": 51963.734375, "p50_us": 52015, "p90_us": 52761, "p95_us": 52848, "p99_us": 52910, "max_us": 52910 }, "inter_token_latency": { "samples": 1984, "min_us": 9997, "mean_us": 11140.15877016129, "p50_us": 11326, "p90_us": 11571, "p95_us": 11630, "p99_us": 11846, "max_us": 12316 }, "time_per_output_token": { "samples": 64, "min_us": 11034, "mean_us": 11140.171875, "p50_us": 11135, "p90_us": 11226, "p95_us": 11229, "p99_us": 11233, "max_us": 11233 }, "end_to_end": { "samples": 64, "min_us": 394008, "mean_us": 397332.140625, "p50_us": 396807, "p90_us": 400656, "p95_us": 400719, "p99_us": 400782, "max_us": 400782 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:55.156599116+00:00", "product": "a3s-gateway", "trial": 4, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.590139746, "streams_per_second": 40.248034904474366, "token_goodput_per_second": 1287.9371169431797, "ttft": { "samples": 64, "min_us": 51003, "mean_us": 51664.234375, "p50_us": 51659, "p90_us": 52161, "p95_us": 52192, "p99_us": 52335, "max_us": 52335 }, "inter_token_latency": { "samples": 1984, "min_us": 10033, "mean_us": 11117.883568548386, "p50_us": 11353, "p90_us": 11611, "p95_us": 11673, "p99_us": 11817, "max_us": 11978 }, "time_per_output_token": { "samples": 64, "min_us": 11016, "mean_us": 11117.890625, "p50_us": 11098, "p90_us": 11304, "p95_us": 11310, "p99_us": 11327, "max_us": 11327 }, "end_to_end": { "samples": 64, "min_us": 392772, "mean_us": 396350.734375, "p50_us": 395726, "p90_us": 402562, "p95_us": 402617, "p99_us": 402981, "max_us": 402981 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:32.069507374+00:00", "product": "a3s-gateway", "trial": 5, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.596302764, "streams_per_second": 40.09264498147545, "token_goodput_per_second": 1282.9646394072145, "ttft": { "samples": 64, "min_us": 50611, "mean_us": 51697.8125, "p50_us": 51696, "p90_us": 52215, "p95_us": 52230, "p99_us": 52390, "max_us": 52390 }, "inter_token_latency": { "samples": 1984, "min_us": 10000, "mean_us": 11162.345262096775, "p50_us": 11332, "p90_us": 11521, "p95_us": 11573, "p99_us": 11861, "max_us": 12120 }, "time_per_output_token": { "samples": 64, "min_us": 10976, "mean_us": 11162.34375, "p50_us": 11142, "p90_us": 11275, "p95_us": 11292, "p99_us": 11307, "max_us": 11307 }, "end_to_end": { "samples": 64, "min_us": 391828, "mean_us": 397754.921875, "p50_us": 397559, "p90_us": 401100, "p95_us": 401692, "p99_us": 401958, "max_us": 401958 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 40.19997765308367, "token_goodput_per_second": 1286.3992848986775, "ttft": { "samples_per_trial": 64, "min_us": 50994, "mean_us": 51825.5, "p50_us": 51853, "p90_us": 52219, "p95_us": 52322, "p99_us": 52410, "max_us": 52410 }, "inter_token_latency": { "samples_per_trial": 1984, "min_us": 10000, "mean_us": 11140.15877016129, "p50_us": 11353, "p90_us": 11600, "p95_us": 11655, "p99_us": 11836, "max_us": 12120 }, "time_per_output_token": { "samples_per_trial": 64, "min_us": 11012, "mean_us": 11140.171875, "p50_us": 11135, "p90_us": 11253, "p95_us": 11254, "p99_us": 11257, "max_us": 11257 }, "end_to_end": { "samples_per_trial": 64, "min_us": 392772, "mean_us": 397332.140625, "p50_us": 397198, "p90_us": 400656, "p95_us": 400747, "p99_us": 400943, "max_us": 400943 }, "cancellation": null, "fault": null } }, "nginx": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:46:57.694654930+00:00", "product": "nginx", "trial": 1, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.589449562, "streams_per_second": 40.26551174072424, "token_goodput_per_second": 1288.4963757031758, "ttft": { "samples": 64, "min_us": 50901, "mean_us": 51633.125, "p50_us": 51573, "p90_us": 52234, "p95_us": 52282, "p99_us": 52342, "max_us": 52342 }, "inter_token_latency": { "samples": 1984, "min_us": 10072, "mean_us": 11126.113407258064, "p50_us": 11343, "p90_us": 11609, "p95_us": 11657, "p99_us": 11724, "max_us": 11920 }, "time_per_output_token": { "samples": 64, "min_us": 10995, "mean_us": 11126.09375, "p50_us": 11130, "p90_us": 11176, "p95_us": 11226, "p99_us": 11240, "max_us": 11240 }, "end_to_end": { "samples": 64, "min_us": 392802, "mean_us": 396559.4375, "p50_us": 396143, "p90_us": 398640, "p95_us": 400081, "p99_us": 400307, "max_us": 400307 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:34.613638706+00:00", "product": "nginx", "trial": 2, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.595526244, "streams_per_second": 40.11215750331463, "token_goodput_per_second": 1283.589040106068, "ttft": { "samples": 64, "min_us": 50559, "mean_us": 51672.265625, "p50_us": 51896, "p90_us": 52136, "p95_us": 52168, "p99_us": 52250, "max_us": 52250 }, "inter_token_latency": { "samples": 1984, "min_us": 8094, "mean_us": 11148.824092741936, "p50_us": 11369, "p90_us": 11531, "p95_us": 11573, "p99_us": 11681, "max_us": 14847 }, "time_per_output_token": { "samples": 64, "min_us": 10998, "mean_us": 11148.859375, "p50_us": 11140, "p90_us": 11238, "p95_us": 11239, "p99_us": 11271, "max_us": 11271 }, "end_to_end": { "samples": 64, "min_us": 391924, "mean_us": 397302.578125, "p50_us": 397236, "p90_us": 400427, "p95_us": 400513, "p99_us": 400882, "max_us": 400882 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:16.045856400+00:00", "product": "nginx", "trial": 3, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.5912069720000002, "streams_per_second": 40.221040459342575, "token_goodput_per_second": 1287.0732946989624, "ttft": { "samples": 64, "min_us": 50467, "mean_us": 51749.484375, "p50_us": 51719, "p90_us": 52276, "p95_us": 52317, "p99_us": 52348, "max_us": 52348 }, "inter_token_latency": { "samples": 1984, "min_us": 10047, "mean_us": 11124.392137096775, "p50_us": 11401, "p90_us": 11643, "p95_us": 11691, "p99_us": 11762, "max_us": 11876 }, "time_per_output_token": { "samples": 64, "min_us": 11009, "mean_us": 11124.359375, "p50_us": 11132, "p90_us": 11205, "p95_us": 11208, "p99_us": 11225, "max_us": 11225 }, "end_to_end": { "samples": 64, "min_us": 393249, "mean_us": 396622.046875, "p50_us": 397079, "p90_us": 399136, "p95_us": 399170, "p99_us": 399566, "max_us": 399566 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:52.907138413+00:00", "product": "nginx", "trial": 4, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.592461123, "streams_per_second": 40.18936417074465, "token_goodput_per_second": 1286.059653463829, "ttft": { "samples": 64, "min_us": 50878, "mean_us": 51874.71875, "p50_us": 52067, "p90_us": 52336, "p95_us": 52351, "p99_us": 52380, "max_us": 52380 }, "inter_token_latency": { "samples": 1984, "min_us": 10024, "mean_us": 11142.098790322581, "p50_us": 11314, "p90_us": 11607, "p95_us": 11647, "p99_us": 11722, "max_us": 12012 }, "time_per_output_token": { "samples": 64, "min_us": 11048, "mean_us": 11142.09375, "p50_us": 11149, "p90_us": 11201, "p95_us": 11247, "p99_us": 11252, "max_us": 11252 }, "end_to_end": { "samples": 64, "min_us": 394139, "mean_us": 397296.59375, "p50_us": 397900, "p90_us": 399490, "p95_us": 400936, "p99_us": 401032, "max_us": 401032 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:34.320136981+00:00", "product": "nginx", "trial": 5, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.5921168909999999, "streams_per_second": 40.19805352344573, "token_goodput_per_second": 1286.3377127502633, "ttft": { "samples": 64, "min_us": 50485, "mean_us": 51527.296875, "p50_us": 51316, "p90_us": 52338, "p95_us": 52435, "p99_us": 52471, "max_us": 52471 }, "inter_token_latency": { "samples": 1984, "min_us": 10033, "mean_us": 11136.961189516129, "p50_us": 11362, "p90_us": 11575, "p95_us": 11620, "p99_us": 11700, "max_us": 11863 }, "time_per_output_token": { "samples": 64, "min_us": 11010, "mean_us": 11136.96875, "p50_us": 11131, "p90_us": 11235, "p95_us": 11236, "p99_us": 11327, "max_us": 11327 }, "end_to_end": { "samples": 64, "min_us": 392489, "mean_us": 396789.71875, "p50_us": 397031, "p90_us": 400580, "p95_us": 400727, "p99_us": 403403, "max_us": 403403 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 40.19805352344573, "token_goodput_per_second": 1286.3377127502633, "ttft": { "samples_per_trial": 64, "min_us": 50559, "mean_us": 51672.265625, "p50_us": 51719, "p90_us": 52276, "p95_us": 52317, "p99_us": 52348, "max_us": 52348 }, "inter_token_latency": { "samples_per_trial": 1984, "min_us": 10033, "mean_us": 11136.961189516129, "p50_us": 11362, "p90_us": 11607, "p95_us": 11647, "p99_us": 11722, "max_us": 11920 }, "time_per_output_token": { "samples_per_trial": 64, "min_us": 11009, "mean_us": 11136.96875, "p50_us": 11132, "p90_us": 11205, "p95_us": 11236, "p99_us": 11252, "max_us": 11252 }, "end_to_end": { "samples_per_trial": 64, "min_us": 392802, "mean_us": 396789.71875, "p50_us": 397079, "p90_us": 399490, "p95_us": 400513, "p99_us": 400882, "max_us": 400882 }, "cancellation": null, "fault": null } } }, "comparison": { "a3s_to_nginx_token_goodput_ratio": 1.0000478662390162, "a3s_to_nginx_stream_rate_ratio": 1.0000478662390162, "a3s_to_nginx_ttft_p50_ratio": 1.0025909240317872, "a3s_to_nginx_ttft_p99_ratio": 1.0011843814472376, "a3s_to_nginx_inter_token_latency_p50_ratio": 0.9992078859355747, "a3s_to_nginx_inter_token_latency_p99_ratio": 1.0097253028493431, "a3s_to_nginx_time_per_output_token_p50_ratio": 1.0002694933524974, "a3s_to_nginx_time_per_output_token_p99_ratio": 1.000444365446143, "a3s_to_nginx_end_to_end_p50_ratio": 1.0002996884750894, "a3s_to_nginx_end_to_end_p99_ratio": 1.0001521644773275, "positions": { "token_goodput": "within_threshold", "stream_rate": "within_threshold", "ttft_p50": "within_threshold", "ttft_p99": "within_threshold", "inter_token_latency_p50": "within_threshold", "inter_token_latency_p99": "within_threshold", "time_per_output_token_p50": "within_threshold", "time_per_output_token_p99": "within_threshold", "end_to_end_p50": "within_threshold", "end_to_end_p99": "within_threshold" } } }, "stream-paced-c64": { "label": "Paced chat stream, C64", "workload": "50 ms first token and 10 ms cadence at concurrency 64", "capability_alignment": "a3s_openai_validation_vs_nginx_transport", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 128, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "products": { "a3s-gateway": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:46:59.161106392+00:00", "product": "a3s-gateway", "trial": 1, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 128, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.802048964, "streams_per_second": 159.59125408208868, "token_goodput_per_second": 5106.920130626838, "ttft": { "samples": 128, "min_us": 51359, "mean_us": 53137.578125, "p50_us": 52978, "p90_us": 54373, "p95_us": 54620, "p99_us": 54832, "max_us": 54886 }, "inter_token_latency": { "samples": 3968, "min_us": 9434, "mean_us": 11172.108114919354, "p50_us": 11047, "p90_us": 11936, "p95_us": 12081, "p99_us": 12336, "max_us": 13508 }, "time_per_output_token": { "samples": 128, "min_us": 11073, "mean_us": 11172.109375, "p50_us": 11177, "p90_us": 11232, "p95_us": 11236, "p99_us": 11247, "max_us": 11248 }, "end_to_end": { "samples": 128, "min_us": 395239, "mean_us": 399493.8515625, "p50_us": 399529, "p90_us": 401151, "p95_us": 401407, "p99_us": 402247, "max_us": 402287 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:39.812488744+00:00", "product": "a3s-gateway", "trial": 2, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 128, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.805252135, "streams_per_second": 158.95642425089628, "token_goodput_per_second": 5086.605576028681, "ttft": { "samples": 128, "min_us": 51144, "mean_us": 52723.4921875, "p50_us": 52848, "p90_us": 54081, "p95_us": 54127, "p99_us": 54250, "max_us": 54340 }, "inter_token_latency": { "samples": 3968, "min_us": 9850, "mean_us": 11223.693548387097, "p50_us": 11078, "p90_us": 12009, "p95_us": 12198, "p99_us": 12708, "max_us": 13216 }, "time_per_output_token": { "samples": 128, "min_us": 11081, "mean_us": 11223.703125, "p50_us": 11231, "p90_us": 11297, "p95_us": 11313, "p99_us": 11327, "max_us": 11340 }, "end_to_end": { "samples": 128, "min_us": 395481, "mean_us": 400678.8203125, "p50_us": 400537, "p90_us": 402756, "p95_us": 403251, "p99_us": 403935, "max_us": 404249 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:17.517640323+00:00", "product": "a3s-gateway", "trial": 3, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 128, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.802098175, "streams_per_second": 159.58146270560957, "token_goodput_per_second": 5106.606806579506, "ttft": { "samples": 128, "min_us": 50864, "mean_us": 52685.4296875, "p50_us": 52320, "p90_us": 53952, "p95_us": 53998, "p99_us": 54100, "max_us": 54105 }, "inter_token_latency": { "samples": 3968, "min_us": 9745, "mean_us": 11162.818044354839, "p50_us": 11073, "p90_us": 11852, "p95_us": 11983, "p99_us": 12256, "max_us": 13573 }, "time_per_output_token": { "samples": 128, "min_us": 10993, "mean_us": 11162.890625, "p50_us": 11145, "p90_us": 11278, "p95_us": 11288, "p99_us": 11324, "max_us": 11325 }, "end_to_end": { "samples": 128, "min_us": 392549, "mean_us": 398753.5390625, "p50_us": 397244, "p90_us": 403418, "p95_us": 403792, "p99_us": 404829, "max_us": 404882 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:58.091626490+00:00", "product": "a3s-gateway", "trial": 4, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 128, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.802882274, "streams_per_second": 159.42561462005824, "token_goodput_per_second": 5101.619667841864, "ttft": { "samples": 128, "min_us": 50961, "mean_us": 52453.9765625, "p50_us": 52418, "p90_us": 53631, "p95_us": 53777, "p99_us": 53887, "max_us": 53897 }, "inter_token_latency": { "samples": 3968, "min_us": 9882, "mean_us": 11187.683215725807, "p50_us": 11002, "p90_us": 11966, "p95_us": 12088, "p99_us": 12248, "max_us": 12534 }, "time_per_output_token": { "samples": 128, "min_us": 11008, "mean_us": 11187.6953125, "p50_us": 11210, "p90_us": 11242, "p95_us": 11245, "p99_us": 11287, "max_us": 11288 }, "end_to_end": { "samples": 128, "min_us": 393763, "mean_us": 399292.6640625, "p50_us": 399897, "p90_us": 400898, "p95_us": 401393, "p99_us": 401997, "max_us": 402005 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:35.793545282+00:00", "product": "a3s-gateway", "trial": 5, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 128, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.805150384, "streams_per_second": 158.97651239274575, "token_goodput_per_second": 5087.248396567864, "ttft": { "samples": 128, "min_us": 50605, "mean_us": 52516.734375, "p50_us": 52592, "p90_us": 53240, "p95_us": 53348, "p99_us": 53476, "max_us": 53787 }, "inter_token_latency": { "samples": 3968, "min_us": 9800, "mean_us": 11181.375252016129, "p50_us": 11057, "p90_us": 11899, "p95_us": 12023, "p99_us": 12188, "max_us": 13225 }, "time_per_output_token": { "samples": 128, "min_us": 11011, "mean_us": 11181.3984375, "p50_us": 11209, "p90_us": 11268, "p95_us": 11271, "p99_us": 11286, "max_us": 11305 }, "end_to_end": { "samples": 128, "min_us": 393815, "mean_us": 399157.78125, "p50_us": 400036, "p90_us": 402098, "p95_us": 402456, "p99_us": 402779, "max_us": 403819 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 159.42561462005824, "token_goodput_per_second": 5101.619667841864, "ttft": { "samples_per_trial": 128, "min_us": 50961, "mean_us": 52685.4296875, "p50_us": 52592, "p90_us": 53952, "p95_us": 53998, "p99_us": 54100, "max_us": 54105 }, "inter_token_latency": { "samples_per_trial": 3968, "min_us": 9800, "mean_us": 11181.375252016129, "p50_us": 11057, "p90_us": 11936, "p95_us": 12081, "p99_us": 12256, "max_us": 13225 }, "time_per_output_token": { "samples_per_trial": 128, "min_us": 11011, "mean_us": 11181.3984375, "p50_us": 11209, "p90_us": 11268, "p95_us": 11271, "p99_us": 11287, "max_us": 11305 }, "end_to_end": { "samples_per_trial": 128, "min_us": 393815, "mean_us": 399292.6640625, "p50_us": 399897, "p90_us": 402098, "p95_us": 402456, "p99_us": 402779, "max_us": 403819 }, "cancellation": null, "fault": null } }, "nginx": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:00.632427560+00:00", "product": "nginx", "trial": 1, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 128, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.806042188, "streams_per_second": 158.8006209918134, "token_goodput_per_second": 5081.619871738028, "ttft": { "samples": 128, "min_us": 51578, "mean_us": 52900.3515625, "p50_us": 52798, "p90_us": 53655, "p95_us": 53793, "p99_us": 54169, "max_us": 54218 }, "inter_token_latency": { "samples": 3968, "min_us": 9583, "mean_us": 11195.638356854839, "p50_us": 11067, "p90_us": 11956, "p95_us": 12080, "p99_us": 12302, "max_us": 12617 }, "time_per_output_token": { "samples": 128, "min_us": 11042, "mean_us": 11195.65625, "p50_us": 11216, "p90_us": 11295, "p95_us": 11297, "p99_us": 11343, "max_us": 11350 }, "end_to_end": { "samples": 128, "min_us": 394680, "mean_us": 399981.7890625, "p50_us": 400715, "p90_us": 403452, "p95_us": 403828, "p99_us": 404488, "max_us": 404759 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:38.333879375+00:00", "product": "nginx", "trial": 2, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 128, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.80467321, "streams_per_second": 159.07078601510793, "token_goodput_per_second": 5090.265152483454, "ttft": { "samples": 128, "min_us": 51187, "mean_us": 52488.125, "p50_us": 52692, "p90_us": 53167, "p95_us": 53309, "p99_us": 53455, "max_us": 53457 }, "inter_token_latency": { "samples": 3968, "min_us": 9798, "mean_us": 11170.124243951614, "p50_us": 11106, "p90_us": 11868, "p95_us": 11980, "p99_us": 12201, "max_us": 12719 }, "time_per_output_token": { "samples": 128, "min_us": 11007, "mean_us": 11170.09375, "p50_us": 11218, "p90_us": 11265, "p95_us": 11293, "p99_us": 11306, "max_us": 11307 }, "end_to_end": { "samples": 128, "min_us": 393058, "mean_us": 398778.359375, "p50_us": 400596, "p90_us": 402140, "p95_us": 403208, "p99_us": 403498, "max_us": 403559 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:18.988816280+00:00", "product": "nginx", "trial": 3, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 128, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.804939838, "streams_per_second": 159.0180954616884, "token_goodput_per_second": 5088.579054774029, "ttft": { "samples": 128, "min_us": 50680, "mean_us": 52313.7109375, "p50_us": 52210, "p90_us": 53317, "p95_us": 53538, "p99_us": 53629, "max_us": 53745 }, "inter_token_latency": { "samples": 3968, "min_us": 9739, "mean_us": 11121.12247983871, "p50_us": 11014, "p90_us": 11790, "p95_us": 11993, "p99_us": 12319, "max_us": 12748 }, "time_per_output_token": { "samples": 128, "min_us": 10976, "mean_us": 11121.1328125, "p50_us": 11106, "p90_us": 11196, "p95_us": 11275, "p99_us": 11302, "max_us": 11304 }, "end_to_end": { "samples": 128, "min_us": 393057, "mean_us": 397084.8515625, "p50_us": 397075, "p90_us": 398994, "p95_us": 401207, "p99_us": 402573, "max_us": 402593 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:56.621784870+00:00", "product": "nginx", "trial": 4, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 128, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.802081185, "streams_per_second": 159.58484302309122, "token_goodput_per_second": 5106.714976738919, "ttft": { "samples": 128, "min_us": 51002, "mean_us": 52429.6953125, "p50_us": 52452, "p90_us": 53250, "p95_us": 53486, "p99_us": 53611, "max_us": 53691 }, "inter_token_latency": { "samples": 3968, "min_us": 10004, "mean_us": 11201.152973790322, "p50_us": 11036, "p90_us": 11952, "p95_us": 12111, "p99_us": 12340, "max_us": 12727 }, "time_per_output_token": { "samples": 128, "min_us": 10981, "mean_us": 11201.140625, "p50_us": 11205, "p90_us": 11289, "p95_us": 11327, "p99_us": 11364, "max_us": 11364 }, "end_to_end": { "samples": 128, "min_us": 391889, "mean_us": 399681.7734375, "p50_us": 399839, "p90_us": 402232, "p95_us": 403392, "p99_us": 403614, "max_us": 403650 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:37.259467485+00:00", "product": "nginx", "trial": 5, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 64, "requests": 128, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 64, "completed_requests": 128, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 4096, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 0.800132798, "streams_per_second": 159.97344480809545, "token_goodput_per_second": 5119.150233859054, "ttft": { "samples": 128, "min_us": 51295, "mean_us": 52733.1015625, "p50_us": 52895, "p90_us": 53596, "p95_us": 53640, "p99_us": 53781, "max_us": 53785 }, "inter_token_latency": { "samples": 3968, "min_us": 9925, "mean_us": 11186.440272177419, "p50_us": 10992, "p90_us": 11921, "p95_us": 12047, "p99_us": 12453, "max_us": 12710 }, "time_per_output_token": { "samples": 128, "min_us": 11063, "mean_us": 11186.453125, "p50_us": 11187, "p90_us": 11231, "p95_us": 11240, "p99_us": 11262, "max_us": 11269 }, "end_to_end": { "samples": 128, "min_us": 395095, "mean_us": 399529.09375, "p50_us": 399530, "p90_us": 401149, "p95_us": 401760, "p99_us": 401974, "max_us": 401984 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 128 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 159.07078601510793, "token_goodput_per_second": 5090.265152483454, "ttft": { "samples_per_trial": 128, "min_us": 51187, "mean_us": 52488.125, "p50_us": 52692, "p90_us": 53317, "p95_us": 53538, "p99_us": 53629, "max_us": 53745 }, "inter_token_latency": { "samples_per_trial": 3968, "min_us": 9798, "mean_us": 11186.440272177419, "p50_us": 11036, "p90_us": 11921, "p95_us": 12047, "p99_us": 12319, "max_us": 12719 }, "time_per_output_token": { "samples_per_trial": 128, "min_us": 11007, "mean_us": 11186.453125, "p50_us": 11205, "p90_us": 11265, "p95_us": 11293, "p99_us": 11306, "max_us": 11307 }, "end_to_end": { "samples_per_trial": 128, "min_us": 393058, "mean_us": 399529.09375, "p50_us": 399839, "p90_us": 402140, "p95_us": 403208, "p99_us": 403498, "max_us": 403559 }, "cancellation": null, "fault": null } } }, "comparison": { "a3s_to_nginx_token_goodput_ratio": 1.0022306333792592, "a3s_to_nginx_stream_rate_ratio": 1.0022306333792592, "a3s_to_nginx_ttft_p50_ratio": 0.9981021786988538, "a3s_to_nginx_ttft_p99_ratio": 1.0087825616737214, "a3s_to_nginx_inter_token_latency_p50_ratio": 1.0019028633562885, "a3s_to_nginx_inter_token_latency_p99_ratio": 0.9948859485347836, "a3s_to_nginx_time_per_output_token_p50_ratio": 1.0003569834895136, "a3s_to_nginx_time_per_output_token_p99_ratio": 0.9983194763842208, "a3s_to_nginx_end_to_end_p50_ratio": 1.0001450583860003, "a3s_to_nginx_end_to_end_p99_ratio": 0.9982180828653425, "positions": { "token_goodput": "within_threshold", "stream_rate": "within_threshold", "ttft_p50": "within_threshold", "ttft_p99": "within_threshold", "inter_token_latency_p50": "within_threshold", "inter_token_latency_p99": "within_threshold", "time_per_output_token_p50": "within_threshold", "time_per_output_token_p99": "within_threshold", "end_to_end_p50": "within_threshold", "end_to_end_p99": "within_threshold" } } }, "stream-long-output": { "label": "Long output stream", "workload": "256-token chat streams for cadence drift and sustained goodput", "capability_alignment": "a3s_openai_validation_vs_nginx_transport", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 32, "token_count": 256, "first_token_delay_ms": 50, "token_interval_ms": 2, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "products": { "a3s-gateway": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:03.455427154+00:00", "product": "a3s-gateway", "trial": 1, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 32, "token_count": 256, "first_token_delay_ms": 50, "token_interval_ms": 2, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.7093176589999999, "streams_per_second": 18.720920498019616, "token_goodput_per_second": 4792.555647493022, "ttft": { "samples": 32, "min_us": 51010, "mean_us": 51925.78125, "p50_us": 51776, "p90_us": 52443, "p95_us": 52477, "p99_us": 52541, "max_us": 52541 }, "inter_token_latency": { "samples": 8160, "min_us": 1771, "mean_us": 3137.3234068627453, "p50_us": 3252, "p90_us": 3402, "p95_us": 3448, "p99_us": 3583, "max_us": 4252 }, "time_per_output_token": { "samples": 32, "min_us": 3118, "mean_us": 3137.375, "p50_us": 3131, "p90_us": 3163, "p95_us": 3163, "p99_us": 3164, "max_us": 3164 }, "end_to_end": { "samples": 32, "min_us": 846719, "mean_us": 852079.4375, "p50_us": 850051, "p90_us": 859110, "p95_us": 859162, "p99_us": 859395, "max_us": 859395 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:45.475440943+00:00", "product": "a3s-gateway", "trial": 2, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 32, "token_count": 256, "first_token_delay_ms": 50, "token_interval_ms": 2, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.712901929, "streams_per_second": 18.68174672363277, "token_goodput_per_second": 4782.527161249989, "ttft": { "samples": 32, "min_us": 50988, "mean_us": 51700.40625, "p50_us": 51504, "p90_us": 52364, "p95_us": 52461, "p99_us": 52496, "max_us": 52496 }, "inter_token_latency": { "samples": 8160, "min_us": 1813, "mean_us": 3141.584926470588, "p50_us": 3256, "p90_us": 3405, "p95_us": 3450, "p99_us": 3549, "max_us": 4019 }, "time_per_output_token": { "samples": 32, "min_us": 3126, "mean_us": 3141.5, "p50_us": 3141, "p90_us": 3154, "p95_us": 3170, "p99_us": 3174, "max_us": 3174 }, "end_to_end": { "samples": 32, "min_us": 849415, "mean_us": 852943.78125, "p50_us": 852613, "p90_us": 856627, "p95_us": 859952, "p99_us": 861945, "max_us": 861945 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:21.795187985+00:00", "product": "a3s-gateway", "trial": 3, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 32, "token_count": 256, "first_token_delay_ms": 50, "token_interval_ms": 2, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.695795787, "streams_per_second": 18.870196662423954, "token_goodput_per_second": 4830.770345580532, "ttft": { "samples": 32, "min_us": 50566, "mean_us": 51769.5, "p50_us": 51957, "p90_us": 52211, "p95_us": 52277, "p99_us": 52296, "max_us": 52296 }, "inter_token_latency": { "samples": 8160, "min_us": 1811, "mean_us": 3115.611519607843, "p50_us": 3243, "p90_us": 3379, "p95_us": 3418, "p99_us": 3504, "max_us": 3918 }, "time_per_output_token": { "samples": 32, "min_us": 3103, "mean_us": 3115.625, "p50_us": 3114, "p90_us": 3125, "p95_us": 3125, "p99_us": 3125, "max_us": 3125 }, "end_to_end": { "samples": 32, "min_us": 842613, "mean_us": 846382.8125, "p50_us": 846275, "p90_us": 848412, "p95_us": 848955, "p99_us": 849009, "max_us": 849009 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:03.736264933+00:00", "product": "a3s-gateway", "trial": 4, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 32, "token_count": 256, "first_token_delay_ms": 50, "token_interval_ms": 2, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.706430069, "streams_per_second": 18.752599699999777, "token_goodput_per_second": 4800.665523199943, "ttft": { "samples": 32, "min_us": 50801, "mean_us": 51536.375, "p50_us": 51519, "p90_us": 52124, "p95_us": 52174, "p99_us": 52210, "max_us": 52210 }, "inter_token_latency": { "samples": 8160, "min_us": 1808, "mean_us": 3130.6237745098038, "p50_us": 3254, "p90_us": 3396, "p95_us": 3436, "p99_us": 3539, "max_us": 4261 }, "time_per_output_token": { "samples": 32, "min_us": 3113, "mean_us": 3130.65625, "p50_us": 3129, "p90_us": 3144, "p95_us": 3149, "p99_us": 3154, "max_us": 3154 }, "end_to_end": { "samples": 32, "min_us": 845085, "mean_us": 849981.53125, "p50_us": 849343, "p90_us": 853927, "p95_us": 855042, "p99_us": 855158, "max_us": 855158 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:40.074584318+00:00", "product": "a3s-gateway", "trial": 5, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 32, "token_count": 256, "first_token_delay_ms": 50, "token_interval_ms": 2, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.702866313, "streams_per_second": 18.791845111801212, "token_goodput_per_second": 4810.71234862111, "ttft": { "samples": 32, "min_us": 50753, "mean_us": 51706.125, "p50_us": 51647, "p90_us": 52278, "p95_us": 52298, "p99_us": 52326, "max_us": 52326 }, "inter_token_latency": { "samples": 8160, "min_us": 1861, "mean_us": 3124.532230392157, "p50_us": 3254, "p90_us": 3394, "p95_us": 3431, "p99_us": 3503, "max_us": 3693 }, "time_per_output_token": { "samples": 32, "min_us": 3104, "mean_us": 3124.5625, "p50_us": 3124, "p90_us": 3136, "p95_us": 3139, "p99_us": 3140, "max_us": 3140 }, "end_to_end": { "samples": 32, "min_us": 842667, "mean_us": 848598.40625, "p50_us": 848918, "p90_us": 852052, "p95_us": 852110, "p99_us": 852222, "max_us": 852222 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 18.752599699999777, "token_goodput_per_second": 4800.665523199943, "ttft": { "samples_per_trial": 32, "min_us": 50801, "mean_us": 51706.125, "p50_us": 51647, "p90_us": 52278, "p95_us": 52298, "p99_us": 52326, "max_us": 52326 }, "inter_token_latency": { "samples_per_trial": 8160, "min_us": 1811, "mean_us": 3130.6237745098038, "p50_us": 3254, "p90_us": 3396, "p95_us": 3436, "p99_us": 3539, "max_us": 4019 }, "time_per_output_token": { "samples_per_trial": 32, "min_us": 3113, "mean_us": 3130.65625, "p50_us": 3129, "p90_us": 3144, "p95_us": 3149, "p99_us": 3154, "max_us": 3154 }, "end_to_end": { "samples_per_trial": 32, "min_us": 845085, "mean_us": 849981.53125, "p50_us": 849343, "p90_us": 853927, "p95_us": 855042, "p99_us": 855158, "max_us": 855158 }, "cancellation": null, "fault": null } }, "nginx": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:06.285102330+00:00", "product": "nginx", "trial": 1, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 32, "token_count": 256, "first_token_delay_ms": 50, "token_interval_ms": 2, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.712809655, "streams_per_second": 18.682753163252105, "token_goodput_per_second": 4782.784809792539, "ttft": { "samples": 32, "min_us": 50582, "mean_us": 51837.03125, "p50_us": 51871, "p90_us": 52281, "p95_us": 52337, "p99_us": 52384, "max_us": 52384 }, "inter_token_latency": { "samples": 8160, "min_us": 1459, "mean_us": 3139.476225490196, "p50_us": 3264, "p90_us": 3413, "p95_us": 3456, "p99_us": 3550, "max_us": 4226 }, "time_per_output_token": { "samples": 32, "min_us": 3107, "mean_us": 3139.46875, "p50_us": 3141, "p90_us": 3168, "p95_us": 3182, "p99_us": 3182, "max_us": 3182 }, "end_to_end": { "samples": 32, "min_us": 843419, "mean_us": 852531.5, "p50_us": 852626, "p90_us": 860251, "p95_us": 863672, "p99_us": 863790, "max_us": 863790 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:42.638911985+00:00", "product": "nginx", "trial": 2, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 32, "token_count": 256, "first_token_delay_ms": 50, "token_interval_ms": 2, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.710900004, "streams_per_second": 18.703606245359506, "token_goodput_per_second": 4788.1231988120335, "ttft": { "samples": 32, "min_us": 50992, "mean_us": 51875.0625, "p50_us": 51790, "p90_us": 52347, "p95_us": 52401, "p99_us": 52459, "max_us": 52459 }, "inter_token_latency": { "samples": 8160, "min_us": 1839, "mean_us": 3138.9029411764704, "p50_us": 3252, "p90_us": 3401, "p95_us": 3445, "p99_us": 3550, "max_us": 4159 }, "time_per_output_token": { "samples": 32, "min_us": 3114, "mean_us": 3138.9375, "p50_us": 3143, "p90_us": 3155, "p95_us": 3155, "p99_us": 3156, "max_us": 3156 }, "end_to_end": { "samples": 32, "min_us": 846555, "mean_us": 852425.21875, "p50_us": 853011, "p90_us": 856485, "p95_us": 856875, "p99_us": 857006, "max_us": 857006 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:24.616943984+00:00", "product": "nginx", "trial": 3, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 32, "token_count": 256, "first_token_delay_ms": 50, "token_interval_ms": 2, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.707721243, "streams_per_second": 18.738421233072405, "token_goodput_per_second": 4797.035835666536, "ttft": { "samples": 32, "min_us": 51182, "mean_us": 51760.65625, "p50_us": 51751, "p90_us": 52115, "p95_us": 52206, "p99_us": 52226, "max_us": 52226 }, "inter_token_latency": { "samples": 8160, "min_us": 1823, "mean_us": 3138.157843137255, "p50_us": 3260, "p90_us": 3395, "p95_us": 3434, "p99_us": 3519, "max_us": 4046 }, "time_per_output_token": { "samples": 32, "min_us": 3100, "mean_us": 3138.09375, "p50_us": 3130, "p90_us": 3152, "p95_us": 3159, "p99_us": 3165, "max_us": 3165 }, "end_to_end": { "samples": 32, "min_us": 842900, "mean_us": 852118.96875, "p50_us": 850299, "p90_us": 855516, "p95_us": 857278, "p99_us": 859027, "max_us": 859027 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:00.915237918+00:00", "product": "nginx", "trial": 4, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 32, "token_count": 256, "first_token_delay_ms": 50, "token_interval_ms": 2, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.710580317, "streams_per_second": 18.707101725642037, "token_goodput_per_second": 4789.018041764361, "ttft": { "samples": 32, "min_us": 50744, "mean_us": 51710.0625, "p50_us": 51869, "p90_us": 52108, "p95_us": 52132, "p99_us": 52190, "max_us": 52190 }, "inter_token_latency": { "samples": 8160, "min_us": 1885, "mean_us": 3144.5270833333334, "p50_us": 3241, "p90_us": 3379, "p95_us": 3418, "p99_us": 3511, "max_us": 4862 }, "time_per_output_token": { "samples": 32, "min_us": 3127, "mean_us": 3144.5, "p50_us": 3148, "p90_us": 3152, "p95_us": 3152, "p99_us": 3152, "max_us": 3152 }, "end_to_end": { "samples": 32, "min_us": 848828, "mean_us": 853691.71875, "p50_us": 854815, "p90_us": 855357, "p95_us": 855394, "p99_us": 855525, "max_us": 855525 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:42.897668345+00:00", "product": "nginx", "trial": 5, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 32, "token_count": 256, "first_token_delay_ms": 50, "token_interval_ms": 2, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 8192, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.714079217, "streams_per_second": 18.66891546354943, "token_goodput_per_second": 4779.242358668654, "ttft": { "samples": 32, "min_us": 51392, "mean_us": 51909.4375, "p50_us": 51956, "p90_us": 52259, "p95_us": 52335, "p99_us": 52380, "max_us": 52380 }, "inter_token_latency": { "samples": 8160, "min_us": 1887, "mean_us": 3144.1174019607843, "p50_us": 3245, "p90_us": 3398, "p95_us": 3445, "p99_us": 3539, "max_us": 5897 }, "time_per_output_token": { "samples": 32, "min_us": 3117, "mean_us": 3144.0625, "p50_us": 3144, "p90_us": 3162, "p95_us": 3170, "p99_us": 3183, "max_us": 3183 }, "end_to_end": { "samples": 32, "min_us": 847128, "mean_us": 853788.0, "p50_us": 853971, "p90_us": 857721, "p95_us": 860049, "p99_us": 863680, "max_us": 863680 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 18.703606245359506, "token_goodput_per_second": 4788.1231988120335, "ttft": { "samples_per_trial": 32, "min_us": 50992, "mean_us": 51837.03125, "p50_us": 51869, "p90_us": 52259, "p95_us": 52335, "p99_us": 52380, "max_us": 52380 }, "inter_token_latency": { "samples_per_trial": 8160, "min_us": 1839, "mean_us": 3139.476225490196, "p50_us": 3252, "p90_us": 3398, "p95_us": 3445, "p99_us": 3539, "max_us": 4226 }, "time_per_output_token": { "samples_per_trial": 32, "min_us": 3114, "mean_us": 3139.46875, "p50_us": 3143, "p90_us": 3155, "p95_us": 3159, "p99_us": 3165, "max_us": 3165 }, "end_to_end": { "samples_per_trial": 32, "min_us": 846555, "mean_us": 852531.5, "p50_us": 853011, "p90_us": 856485, "p95_us": 857278, "p99_us": 859027, "max_us": 859027 }, "cancellation": null, "fault": null } } }, "comparison": { "a3s_to_nginx_token_goodput_ratio": 1.0026194656793754, "a3s_to_nginx_stream_rate_ratio": 1.0026194656793754, "a3s_to_nginx_ttft_p50_ratio": 0.9957199868900499, "a3s_to_nginx_ttft_p99_ratio": 0.9989690721649485, "a3s_to_nginx_inter_token_latency_p50_ratio": 1.0006150061500616, "a3s_to_nginx_inter_token_latency_p99_ratio": 1.0, "a3s_to_nginx_time_per_output_token_p50_ratio": 0.9955456570155902, "a3s_to_nginx_time_per_output_token_p99_ratio": 0.99652448657188, "a3s_to_nginx_end_to_end_p50_ratio": 0.9956999382188506, "a3s_to_nginx_end_to_end_p99_ratio": 0.9954960670619201, "positions": { "token_goodput": "within_threshold", "stream_rate": "within_threshold", "ttft_p50": "within_threshold", "ttft_p99": "within_threshold", "inter_token_latency_p50": "within_threshold", "inter_token_latency_p99": "within_threshold", "time_per_output_token_p50": "within_threshold", "time_per_output_token_p99": "within_threshold", "end_to_end_p50": "within_threshold", "end_to_end_p99": "within_threshold" } } }, "completions-paced-c16": { "label": "Paced completions, C16", "workload": "Legacy completions SSE path with the paced C16 schedule", "capability_alignment": "a3s_openai_validation_vs_nginx_transport", "scenario": { "endpoint": "completions", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "products": { "a3s-gateway": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:08.539072690+00:00", "product": "a3s-gateway", "trial": 1, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "completions", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.5949447719999998, "streams_per_second": 40.12678126763377, "token_goodput_per_second": 1284.0570005642805, "ttft": { "samples": 64, "min_us": 50880, "mean_us": 52189.578125, "p50_us": 52350, "p90_us": 52609, "p95_us": 52649, "p99_us": 52743, "max_us": 52743 }, "inter_token_latency": { "samples": 1984, "min_us": 10039, "mean_us": 11150.023689516129, "p50_us": 11334, "p90_us": 11544, "p95_us": 11602, "p99_us": 11752, "max_us": 12435 }, "time_per_output_token": { "samples": 64, "min_us": 11033, "mean_us": 11150.03125, "p50_us": 11123, "p90_us": 11232, "p95_us": 11233, "p99_us": 11235, "max_us": 11235 }, "end_to_end": { "samples": 64, "min_us": 394582, "mean_us": 397866.625, "p50_us": 397011, "p90_us": 400367, "p95_us": 400716, "p99_us": 400843, "max_us": 400843 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:49.967810270+00:00", "product": "a3s-gateway", "trial": 2, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "completions", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.587931888, "streams_per_second": 40.30399570891419, "token_goodput_per_second": 1289.727862685254, "ttft": { "samples": 64, "min_us": 51009, "mean_us": 51733.640625, "p50_us": 51900, "p90_us": 52169, "p95_us": 52193, "p99_us": 52226, "max_us": 52226 }, "inter_token_latency": { "samples": 1984, "min_us": 9448, "mean_us": 11116.604838709678, "p50_us": 11356, "p90_us": 11567, "p95_us": 11621, "p99_us": 11736, "max_us": 12813 }, "time_per_output_token": { "samples": 64, "min_us": 11052, "mean_us": 11116.609375, "p50_us": 11108, "p90_us": 11189, "p95_us": 11194, "p99_us": 11201, "max_us": 11201 }, "end_to_end": { "samples": 64, "min_us": 394562, "mean_us": 396372.609375, "p50_us": 396173, "p90_us": 398247, "p95_us": 398308, "p99_us": 398390, "max_us": 398390 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:26.873790668+00:00", "product": "a3s-gateway", "trial": 3, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "completions", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.598644934, "streams_per_second": 40.033905365004586, "token_goodput_per_second": 1281.0849716801467, "ttft": { "samples": 64, "min_us": 50474, "mean_us": 51605.65625, "p50_us": 51536, "p90_us": 52247, "p95_us": 52286, "p99_us": 52429, "max_us": 52429 }, "inter_token_latency": { "samples": 1984, "min_us": 10113, "mean_us": 11160.0, "p50_us": 11389, "p90_us": 11560, "p95_us": 11618, "p99_us": 11738, "max_us": 11916 }, "time_per_output_token": { "samples": 64, "min_us": 10996, "mean_us": 11160.03125, "p50_us": 11124, "p90_us": 11284, "p95_us": 11285, "p99_us": 11291, "max_us": 11291 }, "end_to_end": { "samples": 64, "min_us": 392844, "mean_us": 397596.796875, "p50_us": 396261, "p90_us": 401875, "p95_us": 401925, "p99_us": 402045, "max_us": 402045 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:08.239158030+00:00", "product": "a3s-gateway", "trial": 4, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "completions", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.596405436, "streams_per_second": 40.09006644349713, "token_goodput_per_second": 1282.8821261919081, "ttft": { "samples": 64, "min_us": 50740, "mean_us": 51533.046875, "p50_us": 51439, "p90_us": 52321, "p95_us": 52341, "p99_us": 52511, "max_us": 52511 }, "inter_token_latency": { "samples": 1984, "min_us": 10047, "mean_us": 11163.638608870968, "p50_us": 11353, "p90_us": 11558, "p95_us": 11602, "p99_us": 11738, "max_us": 12421 }, "time_per_output_token": { "samples": 64, "min_us": 11032, "mean_us": 11163.625, "p50_us": 11140, "p90_us": 11225, "p95_us": 11256, "p99_us": 11262, "max_us": 11262 }, "end_to_end": { "samples": 64, "min_us": 393712, "mean_us": 397629.046875, "p50_us": 397069, "p90_us": 399807, "p95_us": 400061, "p99_us": 400138, "max_us": 400138 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:45.151429984+00:00", "product": "a3s-gateway", "trial": 5, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "completions", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.594746159, "streams_per_second": 40.13177874034309, "token_goodput_per_second": 1284.2169196909788, "ttft": { "samples": 64, "min_us": 50749, "mean_us": 51642.328125, "p50_us": 51489, "p90_us": 52277, "p95_us": 52327, "p99_us": 52632, "max_us": 52632 }, "inter_token_latency": { "samples": 1984, "min_us": 9497, "mean_us": 11155.560483870968, "p50_us": 11366, "p90_us": 11640, "p95_us": 11689, "p99_us": 11988, "max_us": 16625 }, "time_per_output_token": { "samples": 64, "min_us": 11027, "mean_us": 11155.5, "p50_us": 11158, "p90_us": 11205, "p95_us": 11216, "p99_us": 11218, "max_us": 11218 }, "end_to_end": { "samples": 64, "min_us": 392828, "mean_us": 397500.75, "p50_us": 397235, "p90_us": 399885, "p95_us": 399946, "p99_us": 400034, "max_us": 400034 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 40.12678126763377, "token_goodput_per_second": 1284.0570005642805, "ttft": { "samples_per_trial": 64, "min_us": 50749, "mean_us": 51642.328125, "p50_us": 51536, "p90_us": 52277, "p95_us": 52327, "p99_us": 52511, "max_us": 52511 }, "inter_token_latency": { "samples_per_trial": 1984, "min_us": 10039, "mean_us": 11155.560483870968, "p50_us": 11356, "p90_us": 11560, "p95_us": 11618, "p99_us": 11738, "max_us": 12435 }, "time_per_output_token": { "samples_per_trial": 64, "min_us": 11032, "mean_us": 11155.5, "p50_us": 11124, "p90_us": 11225, "p95_us": 11233, "p99_us": 11235, "max_us": 11235 }, "end_to_end": { "samples_per_trial": 64, "min_us": 393712, "mean_us": 397596.796875, "p50_us": 397011, "p90_us": 399885, "p95_us": 400061, "p99_us": 400138, "max_us": 400138 }, "cancellation": null, "fault": null } }, "nginx": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:10.788694737+00:00", "product": "nginx", "trial": 1, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "completions", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.590963308, "streams_per_second": 40.2272005131623, "token_goodput_per_second": 1287.2704164211937, "ttft": { "samples": 64, "min_us": 50943, "mean_us": 51550.40625, "p50_us": 51486, "p90_us": 51960, "p95_us": 52028, "p99_us": 52216, "max_us": 52216 }, "inter_token_latency": { "samples": 1984, "min_us": 8705, "mean_us": 11150.750504032258, "p50_us": 11347, "p90_us": 11504, "p95_us": 11545, "p99_us": 11618, "max_us": 13314 }, "time_per_output_token": { "samples": 64, "min_us": 11066, "mean_us": 11150.71875, "p50_us": 11154, "p90_us": 11191, "p95_us": 11203, "p99_us": 11205, "max_us": 11205 }, "end_to_end": { "samples": 64, "min_us": 394947, "mean_us": 397240.015625, "p50_us": 397167, "p90_us": 398311, "p95_us": 398371, "p99_us": 399243, "max_us": 399243 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:47.722422315+00:00", "product": "nginx", "trial": 2, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "completions", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.5903905360000001, "streams_per_second": 40.241688158536675, "token_goodput_per_second": 1287.7340210731736, "ttft": { "samples": 64, "min_us": 50898, "mean_us": 51875.046875, "p50_us": 51935, "p90_us": 52372, "p95_us": 52397, "p99_us": 52451, "max_us": 52451 }, "inter_token_latency": { "samples": 1984, "min_us": 10039, "mean_us": 11133.753528225807, "p50_us": 11350, "p90_us": 11611, "p95_us": 11674, "p99_us": 11795, "max_us": 12217 }, "time_per_output_token": { "samples": 64, "min_us": 10993, "mean_us": 11133.6875, "p50_us": 11135, "p90_us": 11199, "p95_us": 11205, "p99_us": 11255, "max_us": 11255 }, "end_to_end": { "samples": 64, "min_us": 392048, "mean_us": 397037.390625, "p50_us": 397081, "p90_us": 398652, "p95_us": 398999, "p99_us": 400859, "max_us": 400859 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:29.126708334+00:00", "product": "nginx", "trial": 3, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "completions", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.595412394, "streams_per_second": 40.115019941358184, "token_goodput_per_second": 1283.680638123462, "ttft": { "samples": 64, "min_us": 50697, "mean_us": 51658.609375, "p50_us": 51689, "p90_us": 52334, "p95_us": 52374, "p99_us": 52437, "max_us": 52437 }, "inter_token_latency": { "samples": 1984, "min_us": 9976, "mean_us": 11172.794858870968, "p50_us": 11372, "p90_us": 11572, "p95_us": 11613, "p99_us": 11749, "max_us": 12209 }, "time_per_output_token": { "samples": 64, "min_us": 10999, "mean_us": 11172.78125, "p50_us": 11173, "p90_us": 11247, "p95_us": 11248, "p99_us": 11304, "max_us": 11304 }, "end_to_end": { "samples": 64, "min_us": 392919, "mean_us": 398031.765625, "p50_us": 398207, "p90_us": 400325, "p95_us": 400489, "p99_us": 401869, "max_us": 401869 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:05.984542843+00:00", "product": "nginx", "trial": 4, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "completions", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.591907998, "streams_per_second": 40.20332838355399, "token_goodput_per_second": 1286.5065082737276, "ttft": { "samples": 64, "min_us": 50898, "mean_us": 51836.234375, "p50_us": 51908, "p90_us": 52349, "p95_us": 52539, "p99_us": 52603, "max_us": 52603 }, "inter_token_latency": { "samples": 1984, "min_us": 10141, "mean_us": 11145.126008064517, "p50_us": 11393, "p90_us": 11552, "p95_us": 11598, "p99_us": 11695, "max_us": 11845 }, "time_per_output_token": { "samples": 64, "min_us": 10999, "mean_us": 11145.0625, "p50_us": 11143, "p90_us": 11222, "p95_us": 11225, "p99_us": 11241, "max_us": 11241 }, "end_to_end": { "samples": 64, "min_us": 393128, "mean_us": 397351.4375, "p50_us": 397541, "p90_us": 399690, "p95_us": 400067, "p99_us": 400565, "max_us": 400565 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:47.401885867+00:00", "product": "nginx", "trial": 5, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "completions", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 1024, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.591467394, "streams_per_second": 40.21445883295301, "token_goodput_per_second": 1286.8626826544962, "ttft": { "samples": 64, "min_us": 50724, "mean_us": 51672.1875, "p50_us": 51605, "p90_us": 52274, "p95_us": 52323, "p99_us": 52386, "max_us": 52386 }, "inter_token_latency": { "samples": 1984, "min_us": 10173, "mean_us": 11128.504536290322, "p50_us": 11417, "p90_us": 11616, "p95_us": 11657, "p99_us": 11752, "max_us": 12005 }, "time_per_output_token": { "samples": 64, "min_us": 11012, "mean_us": 11128.515625, "p50_us": 11120, "p90_us": 11213, "p95_us": 11217, "p99_us": 11257, "max_us": 11257 }, "end_to_end": { "samples": 64, "min_us": 392571, "mean_us": 396672.28125, "p50_us": 396354, "p90_us": 399888, "p95_us": 400008, "p99_us": 401062, "max_us": 401062 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 40.21445883295301, "token_goodput_per_second": 1286.8626826544962, "ttft": { "samples_per_trial": 64, "min_us": 50898, "mean_us": 51672.1875, "p50_us": 51689, "p90_us": 52334, "p95_us": 52374, "p99_us": 52437, "max_us": 52437 }, "inter_token_latency": { "samples_per_trial": 1984, "min_us": 10039, "mean_us": 11145.126008064517, "p50_us": 11372, "p90_us": 11572, "p95_us": 11613, "p99_us": 11749, "max_us": 12209 }, "time_per_output_token": { "samples_per_trial": 64, "min_us": 10999, "mean_us": 11145.0625, "p50_us": 11143, "p90_us": 11213, "p95_us": 11217, "p99_us": 11255, "max_us": 11255 }, "end_to_end": { "samples_per_trial": 64, "min_us": 392919, "mean_us": 397240.015625, "p50_us": 397167, "p90_us": 399690, "p95_us": 400008, "p99_us": 400859, "max_us": 400859 }, "cancellation": null, "fault": null } } }, "comparison": { "a3s_to_nginx_token_goodput_ratio": 0.9978197502126426, "a3s_to_nginx_stream_rate_ratio": 0.9978197502126426, "a3s_to_nginx_ttft_p50_ratio": 0.9970399891659735, "a3s_to_nginx_ttft_p99_ratio": 1.001411217270248, "a3s_to_nginx_inter_token_latency_p50_ratio": 0.998593035525853, "a3s_to_nginx_inter_token_latency_p99_ratio": 0.999063750106392, "a3s_to_nginx_time_per_output_token_p50_ratio": 0.9982948936552095, "a3s_to_nginx_time_per_output_token_p99_ratio": 0.998223011994669, "a3s_to_nginx_end_to_end_p50_ratio": 0.9996072181223515, "a3s_to_nginx_end_to_end_p99_ratio": 0.9982013625738726, "positions": { "token_goodput": "within_threshold", "stream_rate": "within_threshold", "ttft_p50": "within_threshold", "ttft_p99": "within_threshold", "inter_token_latency_p50": "within_threshold", "inter_token_latency_p99": "within_threshold", "time_per_output_token_p50": "within_threshold", "time_per_output_token_p99": "within_threshold", "end_to_end_p50": "within_threshold", "end_to_end_p99": "within_threshold" } } }, "prompt-32k": { "label": "32 KiB prompt", "workload": "32 KiB chat prompt with a paced token stream", "capability_alignment": "a3s_openai_validation_vs_nginx_transport", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 32768, "request_timeout_seconds": 120 }, "products": { "a3s-gateway": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:13.039893855+00:00", "product": "a3s-gateway", "trial": 1, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 32768, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.592924602, "streams_per_second": 40.17767063151932, "token_goodput_per_second": 1285.6854602086182, "ttft": { "samples": 64, "min_us": 50601, "mean_us": 52129.796875, "p50_us": 52114, "p90_us": 52894, "p95_us": 53077, "p99_us": 53208, "max_us": 53208 }, "inter_token_latency": { "samples": 1984, "min_us": 9899, "mean_us": 11145.411794354839, "p50_us": 11319, "p90_us": 11478, "p95_us": 11536, "p99_us": 11655, "max_us": 13363 }, "time_per_output_token": { "samples": 64, "min_us": 11024, "mean_us": 11145.375, "p50_us": 11140, "p90_us": 11206, "p95_us": 11207, "p99_us": 11231, "max_us": 11231 }, "end_to_end": { "samples": 64, "min_us": 394250, "mean_us": 397680.203125, "p50_us": 397427, "p90_us": 399378, "p95_us": 399538, "p99_us": 400474, "max_us": 400474 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:54.471608147+00:00", "product": "a3s-gateway", "trial": 2, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 32768, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.5935834679999998, "streams_per_second": 40.16105920094799, "token_goodput_per_second": 1285.1538944303356, "ttft": { "samples": 64, "min_us": 50861, "mean_us": 51959.15625, "p50_us": 51973, "p90_us": 52638, "p95_us": 52717, "p99_us": 52897, "max_us": 52897 }, "inter_token_latency": { "samples": 1984, "min_us": 10093, "mean_us": 11139.929435483871, "p50_us": 11328, "p90_us": 11597, "p95_us": 11653, "p99_us": 11766, "max_us": 12195 }, "time_per_output_token": { "samples": 64, "min_us": 11019, "mean_us": 11139.9375, "p50_us": 11143, "p90_us": 11208, "p95_us": 11211, "p99_us": 11221, "max_us": 11221 }, "end_to_end": { "samples": 64, "min_us": 392789, "mean_us": 397325.53125, "p50_us": 397337, "p90_us": 399624, "p95_us": 399744, "p99_us": 400202, "max_us": 400202 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:31.374404064+00:00", "product": "a3s-gateway", "trial": 3, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 32768, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.5907760149999999, "streams_per_second": 40.231936738120865, "token_goodput_per_second": 1287.4219756198677, "ttft": { "samples": 64, "min_us": 50731, "mean_us": 52086.96875, "p50_us": 52120, "p90_us": 52992, "p95_us": 53061, "p99_us": 53223, "max_us": 53223 }, "inter_token_latency": { "samples": 1984, "min_us": 9976, "mean_us": 11121.88810483871, "p50_us": 11339, "p90_us": 11519, "p95_us": 11562, "p99_us": 11657, "max_us": 12089 }, "time_per_output_token": { "samples": 64, "min_us": 10989, "mean_us": 11121.96875, "p50_us": 11121, "p90_us": 11195, "p95_us": 11207, "p99_us": 11246, "max_us": 11246 }, "end_to_end": { "samples": 64, "min_us": 392044, "mean_us": 396894.265625, "p50_us": 396559, "p90_us": 399678, "p95_us": 399851, "p99_us": 400750, "max_us": 400750 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:12.736006879+00:00", "product": "a3s-gateway", "trial": 4, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 32768, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.5927643809999998, "streams_per_second": 40.181712225268555, "token_goodput_per_second": 1285.8147912085938, "ttft": { "samples": 64, "min_us": 50481, "mean_us": 52036.515625, "p50_us": 52078, "p90_us": 52966, "p95_us": 53078, "p99_us": 53386, "max_us": 53386 }, "inter_token_latency": { "samples": 1984, "min_us": 10148, "mean_us": 11109.75, "p50_us": 11368, "p90_us": 11566, "p95_us": 11616, "p99_us": 11703, "max_us": 12001 }, "time_per_output_token": { "samples": 64, "min_us": 11028, "mean_us": 11109.734375, "p50_us": 11092, "p90_us": 11224, "p95_us": 11225, "p99_us": 11225, "max_us": 11225 }, "end_to_end": { "samples": 64, "min_us": 393255, "mean_us": 396465.703125, "p50_us": 395387, "p90_us": 400897, "p95_us": 401034, "p99_us": 401367, "max_us": 401367 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:49.654096545+00:00", "product": "a3s-gateway", "trial": 5, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 32768, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.591404909, "streams_per_second": 40.2160378154269, "token_goodput_per_second": 1286.9132100936608, "ttft": { "samples": 64, "min_us": 50834, "mean_us": 52033.5, "p50_us": 51998, "p90_us": 52834, "p95_us": 52938, "p99_us": 53157, "max_us": 53157 }, "inter_token_latency": { "samples": 1984, "min_us": 10198, "mean_us": 11128.129032258064, "p50_us": 11364, "p90_us": 11604, "p95_us": 11655, "p99_us": 11773, "max_us": 12248 }, "time_per_output_token": { "samples": 64, "min_us": 10981, "mean_us": 11128.15625, "p50_us": 11130, "p90_us": 11192, "p95_us": 11202, "p99_us": 11227, "max_us": 11227 }, "end_to_end": { "samples": 64, "min_us": 392576, "mean_us": 397030.90625, "p50_us": 396932, "p90_us": 399162, "p95_us": 399406, "p99_us": 400329, "max_us": 400329 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 40.181712225268555, "token_goodput_per_second": 1285.8147912085938, "ttft": { "samples_per_trial": 64, "min_us": 50731, "mean_us": 52036.515625, "p50_us": 52078, "p90_us": 52894, "p95_us": 53061, "p99_us": 53208, "max_us": 53208 }, "inter_token_latency": { "samples_per_trial": 1984, "min_us": 10093, "mean_us": 11128.129032258064, "p50_us": 11339, "p90_us": 11566, "p95_us": 11616, "p99_us": 11703, "max_us": 12195 }, "time_per_output_token": { "samples_per_trial": 64, "min_us": 11019, "mean_us": 11128.15625, "p50_us": 11130, "p90_us": 11206, "p95_us": 11207, "p99_us": 11227, "max_us": 11227 }, "end_to_end": { "samples_per_trial": 64, "min_us": 392789, "mean_us": 397030.90625, "p50_us": 396932, "p90_us": 399624, "p95_us": 399744, "p99_us": 400474, "max_us": 400474 }, "cancellation": null, "fault": null } }, "nginx": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:15.297008003+00:00", "product": "nginx", "trial": 1, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 32768, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.5968572330000002, "streams_per_second": 40.07872380661346, "token_goodput_per_second": 1282.5191618116307, "ttft": { "samples": 64, "min_us": 50867, "mean_us": 51731.703125, "p50_us": 51666, "p90_us": 52580, "p95_us": 52711, "p99_us": 53079, "max_us": 53079 }, "inter_token_latency": { "samples": 1984, "min_us": 10044, "mean_us": 11148.492943548386, "p50_us": 11361, "p90_us": 11617, "p95_us": 11677, "p99_us": 11758, "max_us": 12125 }, "time_per_output_token": { "samples": 64, "min_us": 11001, "mean_us": 11148.609375, "p50_us": 11142, "p90_us": 11277, "p95_us": 11282, "p99_us": 11290, "max_us": 11290 }, "end_to_end": { "samples": 64, "min_us": 392845, "mean_us": 397351.953125, "p50_us": 396866, "p90_us": 402160, "p95_us": 402465, "p99_us": 402575, "max_us": 402575 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:52.217263180+00:00", "product": "nginx", "trial": 2, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 32768, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.590072425, "streams_per_second": 40.249738938778215, "token_goodput_per_second": 1287.9916460409029, "ttft": { "samples": 64, "min_us": 50875, "mean_us": 51741.515625, "p50_us": 51805, "p90_us": 52201, "p95_us": 52419, "p99_us": 52970, "max_us": 52970 }, "inter_token_latency": { "samples": 1984, "min_us": 9963, "mean_us": 11107.726310483871, "p50_us": 11341, "p90_us": 11611, "p95_us": 11677, "p99_us": 11826, "max_us": 12267 }, "time_per_output_token": { "samples": 64, "min_us": 11045, "mean_us": 11107.859375, "p50_us": 11106, "p90_us": 11195, "p95_us": 11200, "p99_us": 11219, "max_us": 11219 }, "end_to_end": { "samples": 64, "min_us": 393575, "mean_us": 396098.0, "p50_us": 395807, "p90_us": 398915, "p95_us": 399210, "p99_us": 399348, "max_us": 399348 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:33.627797036+00:00", "product": "nginx", "trial": 3, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 32768, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.593238737, "streams_per_second": 40.16974889808996, "token_goodput_per_second": 1285.4319647388788, "ttft": { "samples": 64, "min_us": 50836, "mean_us": 51502.359375, "p50_us": 51424, "p90_us": 52004, "p95_us": 52200, "p99_us": 52645, "max_us": 52645 }, "inter_token_latency": { "samples": 1984, "min_us": 10146, "mean_us": 11160.24697580645, "p50_us": 11359, "p90_us": 11530, "p95_us": 11575, "p99_us": 11680, "max_us": 12201 }, "time_per_output_token": { "samples": 64, "min_us": 11004, "mean_us": 11160.28125, "p50_us": 11161, "p90_us": 11198, "p95_us": 11208, "p99_us": 11231, "max_us": 11231 }, "end_to_end": { "samples": 64, "min_us": 392757, "mean_us": 397486.8125, "p50_us": 397606, "p90_us": 398884, "p95_us": 399316, "p99_us": 399762, "max_us": 399762 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:10.482044987+00:00", "product": "nginx", "trial": 4, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 32768, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.585175864, "streams_per_second": 40.374069182774285, "token_goodput_per_second": 1291.9702138487771, "ttft": { "samples": 64, "min_us": 50347, "mean_us": 51439.265625, "p50_us": 51454, "p90_us": 52132, "p95_us": 52164, "p99_us": 52606, "max_us": 52606 }, "inter_token_latency": { "samples": 1984, "min_us": 10071, "mean_us": 11102.033266129032, "p50_us": 11391, "p90_us": 11576, "p95_us": 11612, "p99_us": 11697, "max_us": 12210 }, "time_per_output_token": { "samples": 64, "min_us": 10982, "mean_us": 11102.046875, "p50_us": 11097, "p90_us": 11179, "p95_us": 11193, "p99_us": 11227, "max_us": 11227 }, "end_to_end": { "samples": 64, "min_us": 392252, "mean_us": 395618.734375, "p50_us": 395429, "p90_us": 398402, "p95_us": 398883, "p99_us": 399181, "max_us": 399181 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:51.908768698+00:00", "product": "nginx", "trial": 5, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 16, "requests": 64, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 32768, "request_timeout_seconds": 120 }, "warmup_requests": 16, "completed_requests": 64, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 2048, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.595690785, "streams_per_second": 40.10802130439075, "token_goodput_per_second": 1283.456681740504, "ttft": { "samples": 64, "min_us": 51191, "mean_us": 51972.3125, "p50_us": 51890, "p90_us": 52517, "p95_us": 52649, "p99_us": 53146, "max_us": 53146 }, "inter_token_latency": { "samples": 1984, "min_us": 10008, "mean_us": 11147.95060483871, "p50_us": 11398, "p90_us": 11589, "p95_us": 11632, "p99_us": 11878, "max_us": 13402 }, "time_per_output_token": { "samples": 64, "min_us": 11025, "mean_us": 11147.9375, "p50_us": 11156, "p90_us": 11225, "p95_us": 11255, "p99_us": 11259, "max_us": 11259 }, "end_to_end": { "samples": 64, "min_us": 393427, "mean_us": 397574.96875, "p50_us": 398080, "p90_us": 400084, "p95_us": 400492, "p99_us": 401361, "max_us": 401361 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 64 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 40.16974889808996, "token_goodput_per_second": 1285.4319647388788, "ttft": { "samples_per_trial": 64, "min_us": 50867, "mean_us": 51731.703125, "p50_us": 51666, "p90_us": 52201, "p95_us": 52419, "p99_us": 52970, "max_us": 52970 }, "inter_token_latency": { "samples_per_trial": 1984, "min_us": 10044, "mean_us": 11147.95060483871, "p50_us": 11361, "p90_us": 11589, "p95_us": 11632, "p99_us": 11758, "max_us": 12210 }, "time_per_output_token": { "samples_per_trial": 64, "min_us": 11004, "mean_us": 11147.9375, "p50_us": 11142, "p90_us": 11198, "p95_us": 11208, "p99_us": 11231, "max_us": 11231 }, "end_to_end": { "samples_per_trial": 64, "min_us": 392845, "mean_us": 397351.953125, "p50_us": 396866, "p90_us": 398915, "p95_us": 399316, "p99_us": 399762, "max_us": 399762 }, "cancellation": null, "fault": null } } }, "comparison": { "a3s_to_nginx_token_goodput_ratio": 1.0002978193169427, "a3s_to_nginx_stream_rate_ratio": 1.0002978193169427, "a3s_to_nginx_ttft_p50_ratio": 1.0079742964425347, "a3s_to_nginx_ttft_p99_ratio": 1.004493109307155, "a3s_to_nginx_inter_token_latency_p50_ratio": 0.9980635507437725, "a3s_to_nginx_inter_token_latency_p99_ratio": 0.9953223337302263, "a3s_to_nginx_time_per_output_token_p50_ratio": 0.9989229940764675, "a3s_to_nginx_time_per_output_token_p99_ratio": 0.9996438429347342, "a3s_to_nginx_end_to_end_p50_ratio": 1.0001663029838788, "a3s_to_nginx_end_to_end_p99_ratio": 1.0017810597305397, "positions": { "token_goodput": "within_threshold", "stream_rate": "within_threshold", "ttft_p50": "within_threshold", "ttft_p99": "within_threshold", "inter_token_latency_p50": "within_threshold", "inter_token_latency_p99": "within_threshold", "time_per_output_token_p50": "within_threshold", "time_per_output_token_p99": "within_threshold", "end_to_end_p50": "within_threshold", "end_to_end_p99": "within_threshold" } } }, "prompt-256k": { "label": "256 KiB prompt", "workload": "256 KiB chat prompt with a paced token stream", "capability_alignment": "a3s_openai_validation_vs_nginx_transport", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 8, "requests": 32, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 262144, "request_timeout_seconds": 120 }, "products": { "a3s-gateway": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:17.561195591+00:00", "product": "a3s-gateway", "trial": 1, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 8, "requests": 32, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 262144, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 1024, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.601357853, "streams_per_second": 19.983041229698205, "token_goodput_per_second": 639.4573193503426, "ttft": { "samples": 32, "min_us": 51623, "mean_us": 53442.84375, "p50_us": 53316, "p90_us": 54760, "p95_us": 55576, "p99_us": 55642, "max_us": 55642 }, "inter_token_latency": { "samples": 992, "min_us": 10097, "mean_us": 11118.464717741936, "p50_us": 11339, "p90_us": 11509, "p95_us": 11547, "p99_us": 11656, "max_us": 11937 }, "time_per_output_token": { "samples": 32, "min_us": 11034, "mean_us": 11118.5, "p50_us": 11117, "p90_us": 11197, "p95_us": 11198, "p99_us": 11214, "max_us": 11214 }, "end_to_end": { "samples": 32, "min_us": 394589, "mean_us": 398137.40625, "p50_us": 398018, "p90_us": 400592, "p95_us": 401514, "p99_us": 401783, "max_us": 401783 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:58.988304658+00:00", "product": "a3s-gateway", "trial": 2, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 8, "requests": 32, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 262144, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 1024, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.59637846, "streams_per_second": 20.045371947702176, "token_goodput_per_second": 641.4519023264696, "ttft": { "samples": 32, "min_us": 51517, "mean_us": 52967.125, "p50_us": 52958, "p90_us": 53763, "p95_us": 53986, "p99_us": 54579, "max_us": 54579 }, "inter_token_latency": { "samples": 992, "min_us": 8847, "mean_us": 11116.568548387097, "p50_us": 11318, "p90_us": 11451, "p95_us": 11503, "p99_us": 11708, "max_us": 12995 }, "time_per_output_token": { "samples": 32, "min_us": 11019, "mean_us": 11116.625, "p50_us": 11119, "p90_us": 11168, "p95_us": 11182, "p99_us": 11183, "max_us": 11183 }, "end_to_end": { "samples": 32, "min_us": 394264, "mean_us": 397608.4375, "p50_us": 397831, "p90_us": 399159, "p95_us": 399437, "p99_us": 399956, "max_us": 399956 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:35.881970259+00:00", "product": "a3s-gateway", "trial": 3, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 8, "requests": 32, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 262144, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 1024, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.592257155, "streams_per_second": 20.097256212361, "token_goodput_per_second": 643.112198795552, "ttft": { "samples": 32, "min_us": 51585, "mean_us": 52557.5625, "p50_us": 52603, "p90_us": 53318, "p95_us": 53733, "p99_us": 53733, "max_us": 53733 }, "inter_token_latency": { "samples": 992, "min_us": 10083, "mean_us": 11097.188508064517, "p50_us": 11312, "p90_us": 11446, "p95_us": 11483, "p99_us": 11559, "max_us": 11796 }, "time_per_output_token": { "samples": 32, "min_us": 11013, "mean_us": 11097.1875, "p50_us": 11103, "p90_us": 11130, "p95_us": 11184, "p99_us": 11185, "max_us": 11185 }, "end_to_end": { "samples": 32, "min_us": 394183, "mean_us": 396660.34375, "p50_us": 396571, "p90_us": 398440, "p95_us": 399327, "p99_us": 400085, "max_us": 400085 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:17.249230133+00:00", "product": "a3s-gateway", "trial": 4, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 8, "requests": 32, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 262144, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 1024, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.601746704, "streams_per_second": 19.978190009748257, "token_goodput_per_second": 639.3020803119442, "ttft": { "samples": 32, "min_us": 51212, "mean_us": 52584.5, "p50_us": 52389, "p90_us": 53525, "p95_us": 54567, "p99_us": 54709, "max_us": 54709 }, "inter_token_latency": { "samples": 992, "min_us": 10117, "mean_us": 11143.630040322581, "p50_us": 11350, "p90_us": 11552, "p95_us": 11608, "p99_us": 11764, "max_us": 12541 }, "time_per_output_token": { "samples": 32, "min_us": 11028, "mean_us": 11143.59375, "p50_us": 11148, "p90_us": 11238, "p95_us": 11242, "p99_us": 11277, "max_us": 11277 }, "end_to_end": { "samples": 32, "min_us": 393672, "mean_us": 398061.875, "p50_us": 398417, "p90_us": 401064, "p95_us": 401649, "p99_us": 401834, "max_us": 401834 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:54.166368724+00:00", "product": "a3s-gateway", "trial": 5, "target": "http://127.0.0.1:18101", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 8, "requests": 32, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 262144, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 1024, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.597812268, "streams_per_second": 20.027384093160563, "token_goodput_per_second": 640.876290981138, "ttft": { "samples": 32, "min_us": 51555, "mean_us": 52927.25, "p50_us": 52670, "p90_us": 55524, "p95_us": 55562, "p99_us": 55612, "max_us": 55612 }, "inter_token_latency": { "samples": 992, "min_us": 10033, "mean_us": 11118.779233870968, "p50_us": 11314, "p90_us": 11552, "p95_us": 11590, "p99_us": 11654, "max_us": 11697 }, "time_per_output_token": { "samples": 32, "min_us": 10987, "mean_us": 11118.8125, "p50_us": 11134, "p90_us": 11175, "p95_us": 11178, "p99_us": 11182, "max_us": 11182 }, "end_to_end": { "samples": 32, "min_us": 393535, "mean_us": 397634.21875, "p50_us": 397041, "p90_us": 401877, "p95_us": 401950, "p99_us": 401963, "max_us": 401963 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 20.027384093160563, "token_goodput_per_second": 640.876290981138, "ttft": { "samples_per_trial": 32, "min_us": 51555, "mean_us": 52927.25, "p50_us": 52670, "p90_us": 53763, "p95_us": 54567, "p99_us": 54709, "max_us": 54709 }, "inter_token_latency": { "samples_per_trial": 992, "min_us": 10083, "mean_us": 11118.464717741936, "p50_us": 11318, "p90_us": 11509, "p95_us": 11547, "p99_us": 11656, "max_us": 11937 }, "time_per_output_token": { "samples_per_trial": 32, "min_us": 11019, "mean_us": 11118.5, "p50_us": 11119, "p90_us": 11175, "p95_us": 11184, "p99_us": 11185, "max_us": 11185 }, "end_to_end": { "samples_per_trial": 32, "min_us": 394183, "mean_us": 397634.21875, "p50_us": 397831, "p90_us": 400592, "p95_us": 401514, "p99_us": 401783, "max_us": 401783 }, "cancellation": null, "fault": null } }, "nginx": { "trials": [ { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:19.815527313+00:00", "product": "nginx", "trial": 1, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 8, "requests": 32, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 262144, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 1024, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.5945809610000001, "streams_per_second": 20.06796818891656, "token_goodput_per_second": 642.1749820453299, "ttft": { "samples": 32, "min_us": 51521, "mean_us": 52485.78125, "p50_us": 52412, "p90_us": 53373, "p95_us": 53395, "p99_us": 53650, "max_us": 53650 }, "inter_token_latency": { "samples": 992, "min_us": 10152, "mean_us": 11115.78125, "p50_us": 11343, "p90_us": 11553, "p95_us": 11577, "p99_us": 11644, "max_us": 14222 }, "time_per_output_token": { "samples": 32, "min_us": 11021, "mean_us": 11115.8125, "p50_us": 11120, "p90_us": 11162, "p95_us": 11174, "p99_us": 11227, "max_us": 11227 }, "end_to_end": { "samples": 32, "min_us": 393523, "mean_us": 397091.9375, "p50_us": 396932, "p90_us": 399196, "p95_us": 400048, "p99_us": 401435, "max_us": 401435 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:47:56.732069881+00:00", "product": "nginx", "trial": 2, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 8, "requests": 32, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 262144, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 1024, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.598222883, "streams_per_second": 20.02223866294123, "token_goodput_per_second": 640.7116372141194, "ttft": { "samples": 32, "min_us": 51586, "mean_us": 52924.53125, "p50_us": 52796, "p90_us": 53626, "p95_us": 54206, "p99_us": 54666, "max_us": 54666 }, "inter_token_latency": { "samples": 992, "min_us": 10032, "mean_us": 11140.183467741936, "p50_us": 11298, "p90_us": 11480, "p95_us": 11526, "p99_us": 11863, "max_us": 12508 }, "time_per_output_token": { "samples": 32, "min_us": 11065, "mean_us": 11140.1875, "p50_us": 11144, "p90_us": 11188, "p95_us": 11209, "p99_us": 11242, "max_us": 11242 }, "end_to_end": { "samples": 32, "min_us": 394619, "mean_us": 398286.625, "p50_us": 398259, "p90_us": 400438, "p95_us": 400956, "p99_us": 401691, "max_us": 401691 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:48:38.133444701+00:00", "product": "nginx", "trial": 3, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 8, "requests": 32, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 262144, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 1024, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.591998088, "streams_per_second": 20.100526653396333, "token_goodput_per_second": 643.2168529086827, "ttft": { "samples": 32, "min_us": 51257, "mean_us": 52549.9375, "p50_us": 52504, "p90_us": 53665, "p95_us": 53703, "p99_us": 54292, "max_us": 54292 }, "inter_token_latency": { "samples": 992, "min_us": 10043, "mean_us": 11118.413306451614, "p50_us": 11297, "p90_us": 11449, "p95_us": 11491, "p99_us": 11610, "max_us": 12287 }, "time_per_output_token": { "samples": 32, "min_us": 11040, "mean_us": 11118.4375, "p50_us": 11116, "p90_us": 11180, "p95_us": 11211, "p99_us": 11218, "max_us": 11218 }, "end_to_end": { "samples": 32, "min_us": 394384, "mean_us": 397237.4375, "p50_us": 397027, "p90_us": 399542, "p95_us": 400974, "p99_us": 401438, "max_us": 401438 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:14.987269998+00:00", "product": "nginx", "trial": 4, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 8, "requests": 32, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 262144, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 1024, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.589909582, "streams_per_second": 20.126930714982006, "token_goodput_per_second": 644.0617828794242, "ttft": { "samples": 32, "min_us": 51298, "mean_us": 52516.25, "p50_us": 52405, "p90_us": 53235, "p95_us": 54198, "p99_us": 54224, "max_us": 54224 }, "inter_token_latency": { "samples": 992, "min_us": 10178, "mean_us": 11087.244959677419, "p50_us": 11367, "p90_us": 11490, "p95_us": 11524, "p99_us": 11637, "max_us": 12383 }, "time_per_output_token": { "samples": 32, "min_us": 11015, "mean_us": 11087.125, "p50_us": 11078, "p90_us": 11147, "p95_us": 11150, "p99_us": 11162, "max_us": 11162 }, "end_to_end": { "samples": 32, "min_us": 394530, "mean_us": 396237.4375, "p50_us": 395793, "p90_us": 397717, "p95_us": 397874, "p99_us": 398712, "max_us": 398712 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] }, { "schema_version": "a3s.gateway.ai-comparison.trial.v1", "generated_at": "2026-09-12T11:49:56.419390954+00:00", "product": "nginx", "trial": 5, "target": "http://127.0.0.1:18102", "scenario": { "endpoint": "chat", "model": "bench", "concurrency": 8, "requests": 32, "token_count": 32, "first_token_delay_ms": 50, "token_interval_ms": 10, "prompt_bytes": 262144, "request_timeout_seconds": 120 }, "warmup_requests": 8, "completed_requests": 32, "failed_requests": 0, "success_rate": 1.0, "completed_tokens": 1024, "intentional_cancels": 0, "intentional_faults": 0, "measured_seconds": 1.594741901, "streams_per_second": 20.065942946588446, "token_goodput_per_second": 642.1101742908303, "ttft": { "samples": 32, "min_us": 51253, "mean_us": 52520.03125, "p50_us": 52443, "p90_us": 53510, "p95_us": 53525, "p99_us": 53620, "max_us": 53620 }, "inter_token_latency": { "samples": 992, "min_us": 10070, "mean_us": 11115.835685483871, "p50_us": 11340, "p90_us": 11470, "p95_us": 11500, "p99_us": 11624, "max_us": 12617 }, "time_per_output_token": { "samples": 32, "min_us": 11026, "mean_us": 11115.9375, "p50_us": 11123, "p90_us": 11164, "p95_us": 11164, "p99_us": 11225, "max_us": 11225 }, "end_to_end": { "samples": 32, "min_us": 393656, "mean_us": 397127.8125, "p50_us": 396938, "p90_us": 399599, "p95_us": 399613, "p99_us": 401495, "max_us": 401495 }, "cancellation": null, "fault": null, "instance_distribution": { "stable": 32 }, "error_samples": [] } ], "median": { "success_rate": 1.0, "streams_per_second": 20.06796818891656, "token_goodput_per_second": 642.1749820453299, "ttft": { "samples_per_trial": 32, "min_us": 51298, "mean_us": 52520.03125, "p50_us": 52443, "p90_us": 53510, "p95_us": 53703, "p99_us": 54224, "max_us": 54224 }, "inter_token_latency": { "samples_per_trial": 992, "min_us": 10070, "mean_us": 11115.835685483871, "p50_us": 11340, "p90_us": 11480, "p95_us": 11524, "p99_us": 11637, "max_us": 12508 }, "time_per_output_token": { "samples_per_trial": 32, "min_us": 11026, "mean_us": 11115.9375, "p50_us": 11120, "p90_us": 11164, "p95_us": 11174, "p99_us": 11225, "max_us": 11225 }, "end_to_end": { "samples_per_trial": 32, "min_us": 394384, "mean_us": 397127.8125, "p50_us": 396938, "p90_us": 399542, "p95_us": 400048, "p99_us": 401438, "max_us": 401438 }, "cancellation": null, "fault": null } } }, "comparison": { "a3s_to_nginx_token_goodput_ratio": 0.9979776679246277, "a3s_to_nginx_stream_rate_ratio": 0.9979776679246277, "a3s_to_nginx_ttft_p50_ratio": 1.0043285090479186, "a3s_to_nginx_ttft_p99_ratio": 1.0089443788728238, "a3s_to_nginx_inter_token_latency_p50_ratio": 0.9980599647266314, "a3s_to_nginx_inter_token_latency_p99_ratio": 1.0016327232104494, "a3s_to_nginx_time_per_output_token_p50_ratio": 0.9999100719424461, "a3s_to_nginx_time_per_output_token_p99_ratio": 0.9964365256124722, "a3s_to_nginx_end_to_end_p50_ratio": 1.0022497216189934, "a3s_to_nginx_end_to_end_p99_ratio": 1.0008594104195419, "positions": { "token_goodput": "within_threshold", "stream_rate": "within_threshold", "ttft_p50": "within_threshold", "ttft_p99": "within_threshold", "inter_token_latency_p50": "within_threshold", "inter_token_latency_p99": "within_threshold", "time_per_output_token_p50": "within_threshold", "time_per_output_token_p99": "within_threshold", "end_to_end_p50": "within_threshold", "end_to_end_p99": "within_threshold" } } } }, "limitations": [ "GitHub-hosted runners are shared regression infrastructure, not a controlled capacity lab.", "The synthetic upstream isolates Gateway transport and request-processing behavior; it does not represent model compute time or quality.", "A3S OpenAI validation and NGINX transport-only forwarding are customer-visible product paths but are not policy-capability equivalent.", "Policy lanes such as telemetry-on report A3S absolute medians and mark NGINX unsupported unless a pinned equivalent module is supplied.", "Cold A3S Box/Sandbox lifecycle and dedicated real-model evidence are separate benchmark lanes." ] }