# helicone vs apisix

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) .

Requested cell: openai → openai. Metric: added_latency_p99_us.

Comparability: comparable. Ratios require an identical protocol pair, hardware, architecture, metric definition, clean harness commit, recorded mock digest, and metric-specific workload. Missing evidence never matches missing evidence. These are observed gateway measurements against a local mock, not a claim about model latency or statistical significance.

- Load-generator binary digests were not recorded; instrument equivalence is based on the identical clean harness source commit and mock digest.
- Measurements were made at different times on independently provisioned machines.

Ratio (B ÷ A): 1.2408759124087592. No statistical significance is asserted.

[Comparison JSON](https://onthebench.ai/api/v1/compare?a=helicone&b=apisix&ingress=openai&egress=openai&metric=added_latency_p99_us)  ·  [All Helicone evidence](https://onthebench.ai/gateways/helicone)  ·  [All APISIX evidence](https://onthebench.ai/gateways/apisix)

## helicone

Build target/release/ai-gateway. Measured 2026-11-10T17:39:37Z.  [Run evidence](https://onthebench.ai/gateways/runs/run-helicone-4a55ff2dcc509c5e03f85bd92e36b949ac410653128ccc77d60019bb3aaf2558)

| Ingress → upstream | Metric | Value | State / reason | p99 bound (ms) |
| --- | --- | --- | --- | --- |
| openai → openai | added_latency_p50_us | 256 | valid; gateway_c1_samples=21418; direct_c1_samples=260967 | — |
| openai → openai | added_latency_p99_us | 274 | valid; gateway_c1_samples=21418; direct_c1_samples=260967 | — |
| openai → openai | gateway_c1_p99_us | 302 | valid; gateway_c1_samples=21418 | — |
| openai → openai | direct_c1_p99_us | 28 | valid; direct_c1_samples=260967 | — |
| openai → openai | cpu_us_per_request | 259.381777553837 | valid | — |
| openai → openai | rps_per_cpu_second | 3855.320945945946 | valid | — |
| openai → openai | cost_window_conc | 8 | valid | — |
| openai → openai | cost_core_utilisation | 0.9840922276022175 | valid | — |
| openai → openai | cost_window_ok | 91294 | valid | — |
| openai → openai | cost_window_rps | 15213 | valid | — |
| openai → openai | cost_threads | 24 | valid | — |
| openai → openai | cost_nonvol_ctxt_per_request | 0 | valid | — |
| openai → openai | cost_majflt | 0 | valid | — |
| openai → openai | added_ttft_p50_us | 399 | valid; ttft_gw_samples=100; ttft_direct_samples=100 | — |
| openai → openai | added_ttft_p99_us | 493 | valid; ttft_gw_samples=100; ttft_direct_samples=100 | — |
| openai → openai | added_gap_p50_us | 0 | valid; gap_gw_samples=63; gap_direct_samples=63 | — |
| openai → openai | added_gap_p99_us | below_resolution | below_resolution; gap_gw_samples=63; gap_direct_samples=63 | — |
| openai → openai | streams_sustained_fps | 21883.075061106818 | valid | — |
| openai → openai | streams_sustained | 463 | valid | — |
| openai → openai | idle_rss_mib | 42.97265625 | valid | — |
| openai → openai | steady_state_rss_mib | 56.1796875 | valid | — |
| openai → openai | recovered_rss_mib | 53.69140625 | valid | — |
| openai → openai | peak_rss_mib | 56.94921875 | valid | — |
| openai → openai | peak_rss_hwm_mib | 56.609375 | valid | — |
| openai → openai | time_to_plateau_s | 126 | valid | — |
| openai → openai | growth_rate_mib_per_min | 0.4999152222931118 | 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 | 0.12448848274528536 | valid | — |
| openai → openai | throughput_rps | 15405 req/s | concurrency 8; samples 92444 | 1 |
| openai → openai | throughput_rps | 16288 req/s | concurrency 32; samples 97741 | 5 |
| openai → openai | throughput_rps | 16376 req/s | concurrency 64; samples 98266 | 10 |
| openai → openai | throughput_rps | 16376 req/s | concurrency 64; samples 98266 | 50 |
| openai → openai | throughput_rps | 16376 req/s | concurrency 64; samples 98266 | 100 |
| openai → openai | throughput_rps | 16376 req/s | concurrency 64; samples 98266 | unbounded |

## apisix

Build apache/apisix:3.19.0-debian@sha256:7401ce7cfd56de1b9dcecfe7d7c7e5a681bc29222767fd28391995a0936eaa38. Measured 2026-11-10T17:39:33Z.  [Run evidence](https://onthebench.ai/gateways/runs/run-apisix-cc145fc7e19bd52b5c47dc4c3db32fb0511f656418b54e198d2b3cf922523d3f)

| Ingress → upstream | Metric | Value | State / reason | p99 bound (ms) |
| --- | --- | --- | --- | --- |
| openai → openai | added_latency_p50_us | 179 | valid; gateway_c1_samples=28247; direct_c1_samples=260315 | — |
| openai → openai | added_latency_p99_us | 340 | valid; gateway_c1_samples=28247; direct_c1_samples=260315 | — |
| openai → openai | gateway_c1_p99_us | 368 | valid; gateway_c1_samples=28247 | — |
| openai → openai | direct_c1_p99_us | 28 | valid; direct_c1_samples=260315 | — |
| openai → openai | cpu_us_per_request | 185.658855916589 | valid | — |
| openai → openai | rps_per_cpu_second | 5386.223000583772 | valid | — |
| openai → openai | cost_window_conc | 8 | valid | — |
| openai → openai | cost_core_utilisation | 0.7114382453326605 | valid | — |
| openai → openai | cost_window_ok | 92266 | valid | — |
| openai → openai | cost_window_rps | 15375 | valid | — |
| openai → openai | cost_threads | 12 | valid | — |
| openai → openai | cost_nonvol_ctxt_per_request | 0.00045520560119654044 | valid | — |
| openai → openai | cost_majflt | 0 | valid | — |
| openai → openai | added_ttft_p50_us | 10584 | valid; ttft_gw_samples=100; ttft_direct_samples=100 | — |
| openai → openai | added_ttft_p99_us | 11228 | 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 | 8877 | valid; gap_gw_samples=63; gap_direct_samples=63 | — |
| openai → openai | streams_sustained_fps | 63882.44005721134 | valid | — |
| openai → openai | streams_sustained | 1470 | valid | — |
| openai → openai | idle_rss_mib | 191.2109375 | valid | — |
| openai → openai | steady_state_rss_mib | 228.22265625 | valid | — |
| openai → openai | recovered_rss_mib | 228.3515625 | valid | — |
| openai → openai | peak_rss_mib | 228.44140625 | valid | — |
| openai → openai | peak_rss_hwm_mib | 233.2109375 | valid | — |
| openai → openai | time_to_plateau_s | 66 | valid | — |
| openai → openai | growth_rate_mib_per_min | 0.050196090921940856 | valid | — |
| openai → openai | load_s | 300 | valid | — |
| openai → openai | shape | not_measured | not_measured | — |
| openai → openai | idle_shape | 0 | valid | — |
| openai → openai | idle_static | 0 | valid | — |
| openai → openai | idle_growth_rate_mib_per_min | -0.46382647366702207 | valid | — |
| openai → openai | throughput_rps | 12187 req/s | concurrency 4; samples 73136 | 1 |
| openai → openai | throughput_rps | 23089 req/s | concurrency 32; samples 138550 | 5 |
| openai → openai | throughput_rps | 25474 req/s | concurrency 64; samples 152851 | 10 |
| openai → openai | throughput_rps | 26590 req/s | concurrency 256; samples 159560 | 50 |
| openai → openai | throughput_rps | 26590 req/s | concurrency 256; samples 159560 | 100 |
| openai → openai | throughput_rps | 26590 req/s | concurrency 256; samples 159560 | unbounded |
