# Kong benchmark measurements

Operated by the Busbar team; Busbar is also an entrant. Gateway overhead against a local mock is not end-to-end model latency.  [Methodology and limitations](https://onthebench.ai/gateways/method) .

Other · recorded build: kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220. Measured: 2026-11-10T17:39:46Z.

[JSON](https://onthebench.ai/api/v1/gateways/kong)  ·  [Markdown](https://onthebench.ai/gateways/kong.md)  ·  [CSV](https://onthebench.ai/gateways/data.csv?gateway=kong&compact=1)  ·  [Project source](https://github.com/Kong/kong)  ·  [Interactive dashboard](https://onthebench.ai/gateways)

Each row below names its protocol pair. A representative dashboard cell may differ between gateways; compare an identical pair before interpreting a ranking. Unavailable metrics retain their reason; below-resolution readings are not exact zeroes.

## Published measurements

| Ingress → upstream | Metric | Value | State / reason | p99 bound (ms) |
| --- | --- | --- | --- | --- |
| cohere → openai | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| bedrock → openai | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| anthropic → openai | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| openai → openai | added_latency_p50_us | 213 | valid; gateway_c1_samples=23861; direct_c1_samples=246038 | — |
| openai → openai | added_latency_p99_us | 410 | valid; gateway_c1_samples=23861; direct_c1_samples=246038 | — |
| openai → openai | gateway_c1_p99_us | 440 | valid; gateway_c1_samples=23861 | — |
| openai → openai | direct_c1_p99_us | 30 | valid; direct_c1_samples=246038 | — |
| openai → openai | cpu_us_per_request | 220.6875669632459 | valid | — |
| openai → openai | rps_per_cpu_second | 4531.292875989446 | valid | — |
| openai → openai | cost_window_conc | 8 | valid | — |
| openai → openai | cost_core_utilisation | 0.787179337229732 | valid | — |
| openai → openai | cost_window_ok | 85868 | valid | — |
| openai → openai | cost_window_rps | 14309 | valid | — |
| openai → openai | cost_threads | 5 | valid | — |
| openai → openai | cost_nonvol_ctxt_per_request | 0.000733684259561187 | valid | — |
| openai → openai | cost_majflt | 0 | valid | — |
| openai → openai | added_ttft_p50_us | 106009 | valid; ttft_gw_samples=100; ttft_direct_samples=100 | — |
| openai → openai | added_ttft_p99_us | 106433 | valid; ttft_gw_samples=100; ttft_direct_samples=100 | — |
| openai → openai | added_gap_p50_us | below_resolution | below_resolution; gap_gw_samples=63; gap_direct_samples=63 | — |
| openai → openai | added_gap_p99_us | 168648 | valid; gap_gw_samples=63; gap_direct_samples=63 | — |
| openai → openai | streams_sustained_fps | 3731.918948013315 | valid | — |
| openai → openai | streams_sustained | 78 | valid | — |
| openai → openai | idle_rss_mib | 406.2109375 | valid | — |
| openai → openai | steady_state_rss_mib | 617.4296875 | valid | — |
| openai → openai | recovered_rss_mib | 617.546875 | valid | — |
| openai → openai | peak_rss_mib | 617.73828125 | valid | — |
| openai → openai | peak_rss_hwm_mib | 618.64453125 | valid | — |
| openai → openai | time_to_plateau_s | 66 | valid | — |
| openai → openai | growth_rate_mib_per_min | 0.10537884976821829 | valid | — |
| openai → openai | load_s | 300 | valid | — |
| openai → openai | shape | not_measured | not_measured | — |
| openai → openai | idle_shape | not_measured | not_measured | — |
| openai → openai | idle_static | 1 | valid | — |
| openai → openai | idle_growth_rate_mib_per_min | 0.240647810533821 | valid | — |
| openai → openai | throughput_rps | 13315 req/s | concurrency 4; samples 79905 | 1 |
| openai → openai | throughput_rps | 19190 req/s | concurrency 8; samples 115157 | 5 |
| openai → openai | throughput_rps | 21287 req/s | concurrency 64; samples 127725 | 10 |
| openai → openai | throughput_rps | 22171 req/s | concurrency 128; samples 133046 | 50 |
| openai → openai | throughput_rps | 22171 req/s | concurrency 128; samples 133046 | 100 |
| openai → openai | throughput_rps | 22171 req/s | concurrency 128; samples 133046 | unbounded |
| gemini → openai | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| openai-responses → openai | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| anthropic → anthropic | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| cohere → anthropic | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| openai → anthropic | added_latency_p50_us | 211 | valid; gateway_c1_samples=24037; direct_c1_samples=241342 | — |
| openai → anthropic | added_latency_p99_us | 420 | valid; gateway_c1_samples=24037; direct_c1_samples=241342 | — |
| openai → anthropic | gateway_c1_p99_us | 450 | valid; gateway_c1_samples=24037 | — |
| openai → anthropic | direct_c1_p99_us | 30 | valid; direct_c1_samples=241342 | — |
| openai → anthropic | cpu_us_per_request | 221.73768782291322 | valid | — |
| openai → anthropic | rps_per_cpu_second | 4509.833262077811 | valid | — |
| openai → anthropic | cost_window_conc | 8 | valid | — |
| openai → anthropic | cost_core_utilisation | 0.9716300019201154 | valid | — |
| openai → anthropic | cost_window_ok | 105485 | valid | — |
| openai → anthropic | cost_window_rps | 17578 | valid | — |
| openai → anthropic | cost_threads | 5 | valid | — |
| openai → anthropic | cost_nonvol_ctxt_per_request | 0.001782243920936626 | valid | — |
| openai → anthropic | cost_majflt | 0 | valid | — |
| openai → anthropic | added_ttft_p50_us | 84892 | valid; ttft_gw_samples=100; ttft_direct_samples=100 | — |
| openai → anthropic | added_ttft_p99_us | 85699 | valid; ttft_gw_samples=100; ttft_direct_samples=100 | — |
| openai → anthropic | added_gap_p50_us | below_resolution | below_resolution; gap_gw_samples=63; gap_direct_samples=63 | — |
| openai → anthropic | added_gap_p99_us | 147628 | valid; gap_gw_samples=63; gap_direct_samples=63 | — |
| openai → anthropic | streams_sustained_fps | 12509.29640828139 | valid | — |
| openai → anthropic | streams_sustained | 264 | valid | — |
| openai → anthropic | idle_rss_mib | 406.140625 | valid | — |
| openai → anthropic | steady_state_rss_mib | 614.625 | valid | — |
| openai → anthropic | recovered_rss_mib | 614.74609375 | valid | — |
| openai → anthropic | peak_rss_mib | 614.87890625 | valid | — |
| openai → anthropic | peak_rss_hwm_mib | 616.01171875 | valid | — |
| openai → anthropic | time_to_plateau_s | 78 | valid | — |
| openai → anthropic | growth_rate_mib_per_min | 0.11096435899246593 | valid | — |
| openai → anthropic | load_s | 300 | valid | — |
| openai → anthropic | shape | not_measured | not_measured | — |
| openai → anthropic | idle_shape | not_measured | not_measured | — |
| openai → anthropic | idle_static | 1 | valid | — |
| openai → anthropic | idle_growth_rate_mib_per_min | 0.32907569010107623 | valid | — |
| openai → anthropic | throughput_rps | 9859 req/s | concurrency 4; samples 59165 | 1 |
| openai → anthropic | throughput_rps | 16975 req/s | concurrency 8; samples 101865 | 5 |
| openai → anthropic | throughput_rps | 18516 req/s | concurrency 64; samples 111100 | 10 |
| openai → anthropic | throughput_rps | 19147 req/s | concurrency 128; samples 114896 | 50 |
| openai → anthropic | throughput_rps | 19147 req/s | concurrency 128; samples 114896 | 100 |
| openai → anthropic | throughput_rps | 19147 req/s | concurrency 128; samples 114896 | unbounded |
| openai-responses → anthropic | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| gemini → anthropic | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| bedrock → anthropic | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| openai → cohere | added_latency_p50_us | 207 | valid; gateway_c1_samples=24758; direct_c1_samples=238237 | — |
| openai → cohere | added_latency_p99_us | 423 | valid; gateway_c1_samples=24758; direct_c1_samples=238237 | — |
| openai → cohere | gateway_c1_p99_us | 454 | valid; gateway_c1_samples=24758 | — |
| openai → cohere | direct_c1_p99_us | 31 | valid; direct_c1_samples=238237 | — |
| openai → cohere | cpu_us_per_request | 216.5942165942166 | valid | — |
| openai → cohere | rps_per_cpu_second | 4616.928446771379 | valid | — |
| openai → cohere | cost_window_conc | 8 | valid | — |
| openai → cohere | cost_core_utilisation | 0.9521305223510478 | valid | — |
| openai → cohere | cost_window_ok | 105820 | valid | — |
| openai → cohere | cost_window_rps | 17634 | valid | — |
| openai → cohere | cost_threads | 5 | valid | — |
| openai → cohere | cost_nonvol_ctxt_per_request | 0.0010584010584010583 | valid | — |
| openai → cohere | cost_majflt | 0 | valid | — |
| openai → cohere | added_ttft_p50_us | untestable | untestable | — |
| openai → cohere | added_ttft_p99_us | untestable | untestable | — |
| openai → cohere | added_gap_p50_us | untestable | untestable | — |
| openai → cohere | added_gap_p99_us | untestable | untestable | — |
| openai → cohere | streams_sustained_fps | untestable | untestable | — |
| openai → cohere | streams_sustained | untestable | untestable | — |
| openai → cohere | idle_rss_mib | 408.06640625 | valid | — |
| openai → cohere | steady_state_rss_mib | 620.25390625 | valid | — |
| openai → cohere | recovered_rss_mib | 620.66015625 | valid | — |
| openai → cohere | peak_rss_mib | 620.78125 | valid | — |
| openai → cohere | peak_rss_hwm_mib | 622.12109375 | valid | — |
| openai → cohere | time_to_plateau_s | 66 | valid | — |
| openai → cohere | growth_rate_mib_per_min | 0.4091809994973029 | valid | — |
| openai → cohere | load_s | 300 | valid | — |
| openai → cohere | shape | not_measured | not_measured | — |
| openai → cohere | idle_shape | not_measured | not_measured | — |
| openai → cohere | idle_static | 1 | valid | — |
| openai → cohere | idle_growth_rate_mib_per_min | 0.2558773134534622 | valid | — |
| openai → cohere | throughput_rps | 12570 req/s | concurrency 4; samples 75430 | 1 |
| openai → cohere | throughput_rps | 18172 req/s | concurrency 16; samples 109045 | 5 |
| openai → cohere | throughput_rps | 18962 req/s | concurrency 32; samples 113777 | 10 |
| openai → cohere | throughput_rps | 20178 req/s | concurrency 128; samples 121088 | 50 |
| openai → cohere | throughput_rps | 20178 req/s | concurrency 128; samples 121088 | 100 |
| openai → cohere | throughput_rps | 20178 req/s | concurrency 128; samples 121088 | unbounded |
| cohere → cohere | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| openai-responses → cohere | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| gemini → cohere | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| bedrock → cohere | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| anthropic → cohere | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| cohere → openai-responses | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| openai → openai-responses | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| anthropic → openai-responses | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| gemini → openai-responses | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| bedrock → openai-responses | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| openai-responses → openai-responses | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| openai → gemini | added_latency_p50_us | 214 | valid; gateway_c1_samples=23733; direct_c1_samples=242637 | — |
| openai → gemini | added_latency_p99_us | 438 | valid; gateway_c1_samples=23733; direct_c1_samples=242637 | — |
| openai → gemini | gateway_c1_p99_us | 468 | valid; gateway_c1_samples=23733 | — |
| openai → gemini | direct_c1_p99_us | 30 | valid; direct_c1_samples=242637 | — |
| openai → gemini | cpu_us_per_request | 214.77521174373956 | valid | — |
| openai → gemini | rps_per_cpu_second | 4656.030795551754 | valid | — |
| openai → gemini | cost_window_conc | 8 | valid | — |
| openai → gemini | cost_core_utilisation | 0.9710935159643339 | valid | — |
| openai → gemini | cost_window_ok | 108858 | valid | — |
| openai → gemini | cost_window_rps | 18140 | valid | — |
| openai → gemini | cost_threads | 5 | valid | — |
| openai → gemini | cost_nonvol_ctxt_per_request | 0.001028863289790369 | valid | — |
| openai → gemini | cost_majflt | 0 | valid | — |
| openai → gemini | added_ttft_p50_us | untestable | untestable | — |
| openai → gemini | added_ttft_p99_us | untestable | untestable | — |
| openai → gemini | added_gap_p50_us | untestable | untestable | — |
| openai → gemini | added_gap_p99_us | untestable | untestable | — |
| openai → gemini | streams_sustained_fps | untestable | untestable | — |
| openai → gemini | streams_sustained | untestable | untestable | — |
| openai → gemini | idle_rss_mib | 408.60546875 | valid | — |
| openai → gemini | steady_state_rss_mib | 622.765625 | valid | — |
| openai → gemini | recovered_rss_mib | 623.0234375 | valid | — |
| openai → gemini | peak_rss_mib | 623.09765625 | valid | — |
| openai → gemini | peak_rss_hwm_mib | 624.20703125 | valid | — |
| openai → gemini | time_to_plateau_s | 66 | valid | — |
| openai → gemini | growth_rate_mib_per_min | 0.26100942658949955 | valid | — |
| openai → gemini | load_s | 300 | valid | — |
| openai → gemini | shape | not_measured | not_measured | — |
| openai → gemini | idle_shape | not_measured | not_measured | — |
| openai → gemini | idle_static | 1 | valid | — |
| openai → gemini | idle_growth_rate_mib_per_min | 0.38746902504991537 | valid | — |
| openai → gemini | throughput_rps | 18455 req/s | concurrency 8; samples 110748 | 1 |
| openai → gemini | throughput_rps | 18455 req/s | concurrency 8; samples 110748 | 5 |
| openai → gemini | throughput_rps | 19059 req/s | concurrency 32; samples 114368 | 10 |
| openai → gemini | throughput_rps | 19713 req/s | concurrency 128; samples 118296 | 50 |
| openai → gemini | throughput_rps | 19713 req/s | concurrency 128; samples 118296 | 100 |
| openai → gemini | throughput_rps | 19713 req/s | concurrency 128; samples 118296 | unbounded |
| openai-responses → gemini | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| cohere → gemini | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| anthropic → gemini | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| gemini → gemini | added_latency_p50_us | 193 | valid; gateway_c1_samples=26272; direct_c1_samples=244214 | — |
| gemini → gemini | added_latency_p99_us | 402 | valid; gateway_c1_samples=26272; direct_c1_samples=244214 | — |
| gemini → gemini | gateway_c1_p99_us | 432 | valid; gateway_c1_samples=26272 | — |
| gemini → gemini | direct_c1_p99_us | 30 | valid; direct_c1_samples=244214 | — |
| gemini → gemini | cpu_us_per_request | 195.4768149651111 | valid | — |
| gemini → gemini | rps_per_cpu_second | 5115.696202531646 | valid | — |
| gemini → gemini | cost_window_conc | 8 | valid | — |
| gemini → gemini | cost_core_utilisation | 0.9844459642780004 | valid | — |
| gemini → gemini | cost_window_ok | 121242 | valid | — |
| gemini → gemini | cost_window_rps | 20204 | valid | — |
| gemini → gemini | cost_threads | 5 | valid | — |
| gemini → gemini | cost_nonvol_ctxt_per_request | 0.0007588129526071824 | valid | — |
| gemini → gemini | cost_majflt | 0 | valid | — |
| gemini → gemini | added_ttft_p50_us | untestable | untestable | — |
| gemini → gemini | added_ttft_p99_us | untestable | untestable | — |
| gemini → gemini | added_gap_p50_us | untestable | untestable | — |
| gemini → gemini | added_gap_p99_us | untestable | untestable | — |
| gemini → gemini | streams_sustained_fps | untestable | untestable | — |
| gemini → gemini | streams_sustained | untestable | untestable | — |
| gemini → gemini | idle_rss_mib | 405.765625 | valid | — |
| gemini → gemini | steady_state_rss_mib | 619.2109375 | valid | — |
| gemini → gemini | recovered_rss_mib | 619.58203125 | valid | — |
| gemini → gemini | peak_rss_mib | 619.78125 | valid | — |
| gemini → gemini | peak_rss_hwm_mib | 621.0234375 | valid | — |
| gemini → gemini | time_to_plateau_s | 66 | valid | — |
| gemini → gemini | growth_rate_mib_per_min | 0.3557036171891926 | valid | — |
| gemini → gemini | load_s | 300 | valid | — |
| gemini → gemini | shape | not_measured | not_measured | — |
| gemini → gemini | idle_shape | not_measured | not_measured | — |
| gemini → gemini | idle_static | 1 | valid | — |
| gemini → gemini | idle_growth_rate_mib_per_min | 0.4672710425108666 | valid | — |
| gemini → gemini | throughput_rps | 13349 req/s | concurrency 4; samples 80104 | 1 |
| gemini → gemini | throughput_rps | 20568 req/s | concurrency 32; samples 123412 | 5 |
| gemini → gemini | throughput_rps | 21659 req/s | concurrency 64; samples 129964 | 10 |
| gemini → gemini | throughput_rps | 21659 req/s | concurrency 64; samples 129964 | 50 |
| gemini → gemini | throughput_rps | 21659 req/s | concurrency 64; samples 129964 | 100 |
| gemini → gemini | throughput_rps | 21659 req/s | concurrency 64; samples 129964 | unbounded |
| bedrock → gemini | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| anthropic → bedrock | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| openai-responses → bedrock | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| cohere → bedrock | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| openai → bedrock | added_latency_p50_us | 1038 | valid; gateway_c1_samples=5167; direct_c1_samples=237310 | — |
| openai → bedrock | added_latency_p99_us | 2133 | valid; gateway_c1_samples=5167; direct_c1_samples=237310 | — |
| openai → bedrock | gateway_c1_p99_us | 2164 | valid; gateway_c1_samples=5167 | — |
| openai → bedrock | direct_c1_p99_us | 31 | valid; direct_c1_samples=237310 | — |
| openai → bedrock | cpu_us_per_request | 1041.3099784889591 | valid | — |
| openai → bedrock | rps_per_cpu_second | 960.3288364249579 | valid | — |
| openai → bedrock | cost_window_conc | 8 | valid | — |
| openai → bedrock | cost_core_utilisation | 0.9846268117121296 | valid | — |
| openai → bedrock | cost_window_ok | 22779 | valid | — |
| openai → bedrock | cost_window_rps | 3796 | valid | — |
| openai → bedrock | cost_threads | 5 | valid | — |
| openai → bedrock | cost_nonvol_ctxt_per_request | 0.003512006672812678 | valid | — |
| openai → bedrock | cost_majflt | 0 | valid | — |
| openai → bedrock | added_ttft_p50_us | untestable | untestable | — |
| openai → bedrock | added_ttft_p99_us | untestable | untestable | — |
| openai → bedrock | added_gap_p50_us | untestable | untestable | — |
| openai → bedrock | added_gap_p99_us | untestable | untestable | — |
| openai → bedrock | streams_sustained_fps | untestable | untestable | — |
| openai → bedrock | streams_sustained | untestable | untestable | — |
| openai → bedrock | idle_rss_mib | 407.8359375 | valid | — |
| openai → bedrock | steady_state_rss_mib | 3411.81640625 | valid | — |
| openai → bedrock | recovered_rss_mib | 3448.46484375 | valid | — |
| openai → bedrock | peak_rss_mib | 3466.08984375 | valid | — |
| openai → bedrock | peak_rss_hwm_mib | 3477.01171875 | valid | — |
| openai → bedrock | time_to_plateau_s | not_measured | not_measured | — |
| openai → bedrock | growth_rate_mib_per_min | 10.902433860925841 | valid | — |
| openai → bedrock | load_s | 300 | valid | — |
| openai → bedrock | shape | 0 | valid | — |
| openai → bedrock | idle_shape | not_measured | not_measured | — |
| openai → bedrock | idle_static | 1 | valid | — |
| openai → bedrock | idle_growth_rate_mib_per_min | 0.3405578567344605 | valid | — |
| openai → bedrock | throughput_rps | bound_not_met req/s | concurrency not recorded; every cleanly-served rung that reported a tail had it at or above 1ms, so this gateway carried no measurable throughput under that bound | 1 |
| openai → bedrock | throughput_rps | 3961 req/s | concurrency 8; samples 23772 | 5 |
| openai → bedrock | throughput_rps | 3961 req/s | concurrency 8; samples 23772 | 10 |
| openai → bedrock | throughput_rps | 3961 req/s | concurrency 8; samples 23772 | 50 |
| openai → bedrock | throughput_rps | 3961 req/s | concurrency 8; samples 23772 | 100 |
| openai → bedrock | throughput_rps | 3961 req/s | concurrency 8; samples 23772 | unbounded |
| gemini → bedrock | — | — | not_declared: Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json | — |
| bedrock → bedrock | added_latency_p50_us | 957 | valid; gateway_c1_samples=5584; direct_c1_samples=241921 | — |
| bedrock → bedrock | added_latency_p99_us | 1988 | valid; gateway_c1_samples=5584; direct_c1_samples=241921 | — |
| bedrock → bedrock | gateway_c1_p99_us | 2018 | valid; gateway_c1_samples=5584 | — |
| bedrock → bedrock | direct_c1_p99_us | 30 | valid; direct_c1_samples=241921 | — |
| bedrock → bedrock | cpu_us_per_request | 1094.7058552964445 | valid | — |
| bedrock → bedrock | rps_per_cpu_second | 913.4873949579832 | valid | — |
| bedrock → bedrock | cost_window_conc | 8 | valid | — |
| bedrock → bedrock | cost_core_utilisation | 0.9885933476042671 | valid | — |
| bedrock → bedrock | cost_window_ok | 21741 | valid | — |
| bedrock → bedrock | cost_window_rps | 3623 | valid | — |
| bedrock → bedrock | cost_threads | 5 | valid | — |
| bedrock → bedrock | cost_nonvol_ctxt_per_request | 0.006669426429327078 | valid | — |
| bedrock → bedrock | cost_majflt | 0 | valid | — |
| bedrock → bedrock | added_ttft_p50_us | untestable | untestable | — |
| bedrock → bedrock | added_ttft_p99_us | untestable | untestable | — |
| bedrock → bedrock | added_gap_p50_us | untestable | untestable | — |
| bedrock → bedrock | added_gap_p99_us | untestable | untestable | — |
| bedrock → bedrock | streams_sustained_fps | untestable | untestable | — |
| bedrock → bedrock | streams_sustained | untestable | untestable | — |
| bedrock → bedrock | idle_rss_mib | 407.671875 | valid | — |
| bedrock → bedrock | steady_state_rss_mib | 3782.19140625 | valid | — |
| bedrock → bedrock | recovered_rss_mib | 3821.51953125 | valid | — |
| bedrock → bedrock | peak_rss_mib | 3840.78515625 | valid | — |
| bedrock → bedrock | peak_rss_hwm_mib | 3862.0625 | valid | — |
| bedrock → bedrock | time_to_plateau_s | not_measured | not_measured | — |
| bedrock → bedrock | growth_rate_mib_per_min | 46.30672272130818 | valid | — |
| bedrock → bedrock | load_s | 300 | valid | — |
| bedrock → bedrock | shape | 0 | valid | — |
| bedrock → bedrock | idle_shape | not_measured | not_measured | — |
| bedrock → bedrock | idle_static | 1 | valid | — |
| bedrock → bedrock | idle_growth_rate_mib_per_min | 0.3255040970998626 | valid | — |
| bedrock → bedrock | throughput_rps | bound_not_met req/s | concurrency not recorded; every cleanly-served rung that reported a tail had it at or above 1ms, so this gateway carried no measurable throughput under that bound | 1 |
| bedrock → bedrock | throughput_rps | 3922 req/s | concurrency 8; samples 23539 | 5 |
| bedrock → bedrock | throughput_rps | 3922 req/s | concurrency 8; samples 23539 | 10 |
| bedrock → bedrock | throughput_rps | 4006 req/s | concurrency 64; samples 24042 | 50 |
| bedrock → bedrock | throughput_rps | 4006 req/s | concurrency 64; samples 24042 | 100 |
| bedrock → bedrock | throughput_rps | 4006 req/s | concurrency 64; samples 24042 | unbounded |

## Probe evidence and instrument provenance

Historical capability notes may reference an earlier product version. They are not a new probe or a current support claim. Full RSS samples remain in the immutable raw JSON.

| Ingress → upstream | Cell state | Probe HTTP status | Recorded reason | Recorded detail | Evidence scope | Response excerpt | Tested build |
| --- | --- | --- | --- | --- | --- | --- | --- |
| cohere → openai | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| bedrock → openai | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| anthropic → openai | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai → openai | served | — | — | — | Recorded run evidence | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| gemini → openai | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai-responses → openai | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| anthropic → anthropic | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| cohere → anthropic | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai → anthropic | served | — | — | — | Recorded run evidence | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai-responses → anthropic | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| gemini → anthropic | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| bedrock → anthropic | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai → cohere | served | — | — | — | Recorded run evidence | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| cohere → cohere | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai-responses → cohere | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| gemini → cohere | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| bedrock → cohere | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| anthropic → cohere | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| cohere → openai-responses | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai → openai-responses | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| anthropic → openai-responses | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| gemini → openai-responses | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| bedrock → openai-responses | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai-responses → openai-responses | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai → gemini | served | — | — | — | Recorded run evidence | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai-responses → gemini | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| cohere → gemini | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| anthropic → gemini | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| gemini → gemini | served | — | — | — | Recorded run evidence | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| bedrock → gemini | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| anthropic → bedrock | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai-responses → bedrock | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| cohere → bedrock | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| openai → bedrock | served | — | — | — | Recorded run evidence | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| gemini → bedrock | not_declared | — | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 1](https://onthebench.ai/#note-1) | [recorded note 2](https://onthebench.ai/#note-2) | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |
| bedrock → bedrock | served | — | — | — | Recorded run evidence | — | kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220 |

**Harness commit**: `d240badd966933fa19eec0606b2d1f3945f18ad2`

**Mock SHA-256**: `99d01f709d8d9c1f49055aa265d5eb7475264d6b513911fe317aee74a71190f8`

**Load generator SHA-256**: `not recorded`

**somaxconn**: 65535

**tcp_max_syn_backlog**: 65535

## Stream sweep diagnostics

Per streaming-sweep concurrency, as recorded by the engine: lanes by final HTTP status, and lanes that received no response.

| Ingress → upstream | Concurrency | Lanes by HTTP status | Lanes with no response |
| --- | --- | --- | --- |
| openai → openai | 1 | {"200":1} | 0 |
| openai → openai | 2 | {"200":2} | 0 |
| openai → openai | 4 | {"200":4} | 0 |
| openai → openai | 8 | {"200":8} | 0 |
| openai → openai | 16 | {"200":16} | 0 |
| openai → openai | 32 | {"200":32} | 0 |
| openai → openai | 64 | {"200":64} | 0 |
| openai → openai | 128 | {"200":128} | 0 |
| openai → openai | 96 | {"200":96} | 0 |
| openai → openai | 80 | {"200":80} | 0 |
| openai → openai | 72 | {"200":72} | 0 |
| openai → openai | 76 | {"200":76} | 0 |
| openai → openai | 78 | {"200":78} | 0 |
| openai → openai | 79 | {"200":79} | 0 |
| openai → openai | 78 | {"200":78} | 0 |
| openai → openai | 78 | {"200":78} | 0 |
| openai → anthropic | 1 | {"200":1} | 0 |
| openai → anthropic | 2 | {"200":2} | 0 |
| openai → anthropic | 4 | {"200":4} | 0 |
| openai → anthropic | 8 | {"200":8} | 0 |
| openai → anthropic | 16 | {"200":16} | 0 |
| openai → anthropic | 32 | {"200":32} | 0 |
| openai → anthropic | 64 | {"200":64} | 0 |
| openai → anthropic | 128 | {"200":128} | 0 |
| openai → anthropic | 256 | {"200":256} | 0 |
| openai → anthropic | 512 | {"200":512} | 0 |
| openai → anthropic | 384 | {"200":384} | 0 |
| openai → anthropic | 320 | {"200":320} | 0 |
| openai → anthropic | 288 | {"200":288} | 0 |
| openai → anthropic | 272 | {"200":272} | 0 |
| openai → anthropic | 280 | {"200":280} | 0 |
| openai → anthropic | 276 | {"200":276} | 0 |
| openai → anthropic | 274 | {"200":274} | 0 |
| openai → anthropic | 273 | {"200":273} | 0 |
| openai → anthropic | 272 | {"200":272} | 0 |
| openai → anthropic | 272 | {"200":272} | 0 |
| openai → anthropic | 264 | {"200":264} | 0 |
| openai → anthropic | 264 | {"200":264} | 0 |
| openai → anthropic | 264 | {"200":264} | 0 |

## Recorded notes

Text recorded with the cells above. Each note is shown once; rows refer to it by number.

- Pinned Kong 3.9.4 ai-proxy schema supports OpenAI ingress plus native Gemini and Bedrock llm_format, with native formats restricted to their matching provider. OpenAI ingress is configured for five providers including Cohere v1. Seven exact pairs passed local response and upstream-recorder preflight. OpenAI Responses and native Anthropic/Cohere ingress are not exposed by this ai-proxy schema. Evidence: results/diagnostics/refresh-configuration/kong-protocol-preflight.json
- Historical capability research, not a fresh probe of the tested build. Referenced versions may differ from gateway_build; do not infer current capability.

## Evidence for the current measurements

- [run-kong-d4b6d413ea59fd281dacf70be5988600ace51a15e8665c6c7dedcc87dd2cd2fb](https://onthebench.ai/gateways/runs/run-kong-d4b6d413ea59fd281dacf70be5988600ace51a15e8665c6c7dedcc87dd2cd2fb)  ·  [complete raw record including RSS series](https://onthebench.ai/api/v1/runs/run-kong-d4b6d413ea59fd281dacf70be5988600ace51a15e8665c6c7dedcc87dd2cd2fb/raw.json)

## Configuration recorded by the run

```
_format_version: "3.0"
# ONE static config, five upstream providers and two native ingress routes. The client picks the upstream with a request header on
# the SAME uniform OpenAI path: \`x-llm-provider: anthropic|gemini|bedrock\`; no header = openai.
# Route priority puts every header-matched route above the header-less openai fallback
# (kong/router/transform.lua get_priority: match_weight occupies the top 3 bits).
services:
  - name: llm
    # Placeholder: ai-proxy overrides host/port/scheme/path per plugin instance via
    # kong.service.set_target() + set_path() from model.options.upstream_url.
    url: http://127.0.0.1:1
    routes:
      - name: chat-anthropic
        paths: ["/v1/chat/completions"]
        headers:
          x-llm-provider: ["anthropic"]
        strip_path: false
        plugins:
          - name: ai-proxy
            config:
              route_type: llm/v1/chat
              auth:
                header_name: Authorization
                header_value: "Bearer dummy"
              model:
                provider: anthropic
                name: gpt-4o-mini
                options:
                  anthropic_version: "2023-06-01"
                  upstream_url: "http://127.0.0.1:8000/v1/messages"
      - name: chat-gemini
        paths: ["/v1/chat/completions"]
        headers:
          x-llm-provider: ["gemini"]
        strip_path: false
        plugins:
          - name: ai-proxy
            config:
              route_type: llm/v1/chat
              auth:
                header_name: Authorization
                header_value: "Bearer dummy"
              model:
                provider: gemini
                name: gpt-4o-mini
                options:
                  upstream_url: "http://127.0.0.1:8000/v1beta/models/gpt-4o-mini:generateContent"
      - name: chat-bedrock
        paths: ["/v1/chat/completions"]
        headers:
          x-llm-provider: ["bedrock"]
        strip_path: false
        plugins:
          - name: ai-proxy
            config:
              route_type: llm/v1/chat
              auth:
                aws_access_key_id: "AKIAMOCKACCESSKEY"
                aws_secret_access_key: "mock-secret-access-key"
              model:
                provider: bedrock
                name: gpt-4o-mini
                options:
                  bedrock:
                    aws_region: "us-east-1"
                  upstream_url: "http://127.0.0.1:8000/model/gpt-4o-mini/converse"
      - name: chat-cohere
        paths: ["/v1/chat/completions"]
        headers:
          x-llm-provider: ["cohere"]
        strip_path: false
        plugins:
          - name: ai-proxy
            config:
              route_type: llm/v1/chat
              auth:
                header_name: Authorization
                header_value: "Bearer dummy"
              model:
                provider: cohere
                name: gpt-4o-mini
                options:
                  upstream_url: "http://127.0.0.1:8000/v1/chat"
      - name: native-gemini
        paths: ["/v1beta/models/gpt-4o-mini:generateContent"]
        strip_path: false
        plugins:
          - name: ai-proxy
            config:
              route_type: llm/v1/chat
              llm_format: gemini
              auth:
                header_name: Authorization
                header_value: "Bearer dummy"
              model:
                provider: gemini
                name: gpt-4o-mini
                options:
                  upstream_url: "http://127.0.0.1:8000/v1beta/models/gpt-4o-mini:generateContent"
      - name: native-bedrock
        paths: ["/model/gpt-4o-mini/converse"]
        strip_path: false
        plugins:
          - name: ai-proxy
            config:
              route_type: llm/v1/chat
              llm_format: bedrock
              auth:
                aws_access_key_id: "AKIAMOCKACCESSKEY"
                aws_secret_access_key: "mock-secret-access-key"
              model:
                provider: bedrock
                name: gpt-4o-mini
                options:
                  bedrock:
                    aws_region: "us-east-1"
                  upstream_url: "http://127.0.0.1:8000/model/gpt-4o-mini/converse"
      # Header-less fallback: the OOTB OpenAI client experience (lowest route priority).
      - name: chat
        paths: ["/v1/chat/completions"]
        strip_path: false
        plugins:
          - name: ai-proxy
            config:
              route_type: llm/v1/chat
              auth:
                header_name: Authorization
                header_value: "Bearer dummy"
              model:
                provider: openai
                name: gpt-4o-mini
                options:
                  upstream_url: "http://127.0.0.1:8000/v1/chat/completions"

```

## Compare a shared protocol cell

- [Kong vs agentgateway](https://onthebench.ai/gateways/compare/kong-vs-agentgateway)
- [Kong vs AISIX (api7)](https://onthebench.ai/gateways/compare/kong-vs-aisix)
- [Kong vs APISIX](https://onthebench.ai/gateways/compare/kong-vs-apisix)
- [Kong vs Bifrost](https://onthebench.ai/gateways/compare/kong-vs-bifrost)
- [Kong vs Busbar](https://onthebench.ai/gateways/compare/kong-vs-busbar)
- [Kong vs GoModel](https://onthebench.ai/gateways/compare/kong-vs-gomodel)
- [Kong vs Helicone](https://onthebench.ai/gateways/compare/kong-vs-helicone)
- [Kong vs Higress](https://onthebench.ai/gateways/compare/kong-vs-higress)
- [Kong vs LiteLLM · Python](https://onthebench.ai/gateways/compare/kong-vs-litellm-python)
- [Kong vs LiteLLM · Rust](https://onthebench.ai/gateways/compare/kong-vs-litellm-rust)
- [Kong vs One-API](https://onthebench.ai/gateways/compare/kong-vs-one-api)
- [Kong vs Plano](https://onthebench.ai/gateways/compare/kong-vs-plano)
- [Kong vs Portkey](https://onthebench.ai/gateways/compare/kong-vs-portkey)
- [Kong vs TensorZero](https://onthebench.ai/gateways/compare/kong-vs-tensorzero)

## Historical measurements

- [2026-11-10T17:39:46Z — kong:3.9.4@sha256:2d572848be6fc58b73ba75edc84f028837bb7be998b3a8b8117fdb44a6892220](https://onthebench.ai/gateways/runs/run-kong-d4b6d413ea59fd281dacf70be5988600ace51a15e8665c6c7dedcc87dd2cd2fb)
- [2026-08-03T22:17:28Z — kong:3.9.3](https://onthebench.ai/gateways/runs/run-kong-15ab99f39305beb017cb61146e45e6036349ad37ae66ee0d59cdd0064fe4e7d5)
