# bifrost vs litellm-rust

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): 0.2639545884578997. No statistical significance is asserted.

[Comparison JSON](https://onthebench.ai/api/v1/compare?a=bifrost&b=litellm-rust&ingress=openai&egress=openai&metric=added_latency_p99_us)  ·  [All Bifrost evidence](https://onthebench.ai/gateways/bifrost)  ·  [All LiteLLM · Rust evidence](https://onthebench.ai/gateways/litellm-rust)

## bifrost

Build maximhq/bifrost:v2.5.0@sha256:445bc49544bc64971b63581a93c6ca5716a03ccf27a58ced4c285cbafc133e78. Measured 2026-11-11T03:17:59Z.  [Run evidence](https://onthebench.ai/gateways/runs/run-bifrost-c253d4668350f1793f587661b425702d5448ca78d04ec9144589514b6064ebd7)

| Ingress → upstream | Metric | Value | State / reason | p99 bound (ms) |
| --- | --- | --- | --- | --- |
| openai → openai | added_latency_p50_us | 658 | valid; gateway_c1_samples=8543; direct_c1_samples=269310 | — |
| openai → openai | added_latency_p99_us | 1057 | valid; gateway_c1_samples=8543; direct_c1_samples=269310 | — |
| openai → openai | gateway_c1_p99_us | 1084 | valid; gateway_c1_samples=8543 | — |
| openai → openai | direct_c1_p99_us | 27 | valid; direct_c1_samples=269310 | — |
| openai → openai | cpu_us_per_request | 865.1196519216824 | valid | — |
| openai → openai | rps_per_cpu_second | 1155.9094719195307 | valid | — |
| openai → openai | cost_window_conc | 8 | valid | — |
| openai → openai | cost_core_utilisation | 0.9909277682234482 | valid | — |
| openai → openai | cost_window_ok | 27580 | valid | — |
| openai → openai | cost_window_rps | 4595 | valid | — |
| openai → openai | cost_threads | 13 | valid | — |
| openai → openai | cost_nonvol_ctxt_per_request | 0.02708484408992023 | valid | — |
| openai → openai | cost_majflt | 0 | valid | — |
| openai → openai | added_ttft_p50_us | 932 | valid; ttft_gw_samples=100; ttft_direct_samples=100 | — |
| openai → openai | added_ttft_p99_us | 1417 | valid; ttft_gw_samples=100; ttft_direct_samples=100 | — |
| openai → openai | added_gap_p50_us | 3 | 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 | 25840.89500725109 | valid | — |
| openai → openai | streams_sustained | 580 | valid | — |
| openai → openai | idle_rss_mib | 160.2734375 | valid | — |
| openai → openai | steady_state_rss_mib | 1049.515625 | valid | — |
| openai → openai | recovered_rss_mib | 1256.6328125 | valid | — |
| openai → openai | peak_rss_mib | 1257.0390625 | valid | — |
| openai → openai | peak_rss_hwm_mib | 1137.43359375 | valid | — |
| openai → openai | time_to_plateau_s | not_measured | not_measured | — |
| openai → openai | growth_rate_mib_per_min | 56.82213581346736 | valid | — |
| openai → openai | load_s | 300 | valid | — |
| openai → openai | shape | 1 | valid | — |
| openai → openai | idle_shape | not_measured | not_measured | — |
| openai → openai | idle_static | 1 | valid | — |
| openai → openai | idle_growth_rate_mib_per_min | 0.4802766176196459 | valid | — |
| openai → openai | 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 → openai | throughput_rps | 4346 req/s | concurrency 4; samples 26081 | 5 |
| openai → openai | throughput_rps | 4555 req/s | concurrency 8; samples 27340 | 10 |
| openai → openai | throughput_rps | 4624 req/s | concurrency 16; samples 27748 | 50 |
| openai → openai | throughput_rps | 4624 req/s | concurrency 16; samples 27748 | 100 |
| openai → openai | throughput_rps | 4624 req/s | concurrency 16; samples 27748 | unbounded |

## litellm-rust

Build src/litellm-rust/target/release/litellm-gateway. Measured 2026-11-10T17:39:51Z.  [Run evidence](https://onthebench.ai/gateways/runs/run-litellm-rust-69d8e6ed325740d9da7f0aaf31b815418c4801eb6549b2e2c68f768a92af3ad4)

| Ingress → upstream | Metric | Value | State / reason | p99 bound (ms) |
| --- | --- | --- | --- | --- |
| openai → openai | added_latency_p50_us | 249 | valid; gateway_c1_samples=22021; direct_c1_samples=267021 | — |
| openai → openai | added_latency_p99_us | 279 | valid; gateway_c1_samples=22021; direct_c1_samples=267021 | — |
| openai → openai | gateway_c1_p99_us | 306 | valid; gateway_c1_samples=22021 | — |
| openai → openai | direct_c1_p99_us | 27 | valid; direct_c1_samples=267021 | — |
| openai → openai | cpu_us_per_request | 272.6013913681787 | valid | — |
| openai → openai | rps_per_cpu_second | 3668.359853121175 | valid | — |
| openai → openai | cost_window_conc | 8 | valid | — |
| openai → openai | cost_core_utilisation | 0.6790579018170796 | valid | — |
| openai → openai | cost_window_ok | 59941 | valid | — |
| openai → openai | cost_window_rps | 9989 | valid | — |
| openai → openai | cost_threads | 5 | valid | — |
| openai → openai | cost_nonvol_ctxt_per_request | 0 | valid | — |
| openai → openai | cost_majflt | 0 | valid | — |
| openai → openai | added_ttft_p50_us | not_measured | not_measured | — |
| openai → openai | added_ttft_p99_us | not_measured | not_measured | — |
| openai → openai | added_gap_p50_us | not_measured | not_measured | — |
| openai → openai | added_gap_p99_us | not_measured | not_measured | — |
| openai → openai | streams_sustained_fps | 0 (measured_failure) | valid | — |
| openai → openai | streams_sustained | 0 (measured_failure) | valid | — |
| openai → openai | idle_rss_mib | 12.73828125 | valid | — |
| openai → openai | steady_state_rss_mib | 18.9921875 | valid | — |
| openai → openai | recovered_rss_mib | 18.99609375 | valid | — |
| openai → openai | peak_rss_mib | 19.875 | valid | — |
| openai → openai | peak_rss_hwm_mib | 19.484375 | valid | — |
| openai → openai | time_to_plateau_s | not_measured | not_measured | — |
| openai → openai | growth_rate_mib_per_min | -0.1791634760562065 | valid | — |
| openai → openai | load_s | 300 | valid | — |
| openai → openai | shape | 0 | valid | — |
| openai → openai | idle_shape | not_measured | not_measured | — |
| openai → openai | idle_static | 1 | valid | — |
| openai → openai | idle_growth_rate_mib_per_min | 0 | valid | — |
| openai → openai | throughput_rps | 8404 req/s | concurrency 4; samples 50435 | 1 |
| openai → openai | throughput_rps | 10049 req/s | concurrency 16; samples 60306 | 5 |
| openai → openai | throughput_rps | 10049 req/s | concurrency 16; samples 60306 | 10 |
| openai → openai | throughput_rps | 10049 req/s | concurrency 16; samples 60306 | 50 |
| openai → openai | throughput_rps | 10049 req/s | concurrency 16; samples 60306 | 100 |
| openai → openai | throughput_rps | 10049 req/s | concurrency 16; samples 60306 | unbounded |
