# On the Bench > Reproducible AI gateway measurements. Operated by the Busbar team, also an entrant. Publication generated 2026-11-14T11:09:24.858Z; newest measurement 2026-11-11T03:17:59Z. Independent runs, not one simultaneous field experiment. ## Resources - [Gateway benchmarks](https://onthebench.ai/gateways) - [Markdown results](https://onthebench.ai/gateways/performance.md) - [Data and API documentation](https://onthebench.ai/gateways/data) - [Dataset reuse license](https://onthebench.ai/gateways/license) - [Compact summary — start here](https://onthebench.ai/api/v1/summary) - [Gateway catalog](https://onthebench.ai/api/v1/gateways) - [API discovery](https://onthebench.ai/api/v1) - [Latest dataset — large; prefer summary or paginated measurements](https://onthebench.ai/api/v1/benchmarks/latest) - [Methodology](https://onthebench.ai/gateways/method.md) - [Historical runs](https://onthebench.ai/gateways/runs) - [OpenAPI](https://onthebench.ai/openapi.json) ## Gateways - [agentgateway](https://onthebench.ai/gateways/agentgateway.md) - [AISIX (api7)](https://onthebench.ai/gateways/aisix.md) - [APISIX](https://onthebench.ai/gateways/apisix.md) - [Bifrost](https://onthebench.ai/gateways/bifrost.md) - [Busbar](https://onthebench.ai/gateways/busbar.md) - [GoModel](https://onthebench.ai/gateways/gomodel.md) - [Helicone](https://onthebench.ai/gateways/helicone.md) - [Higress](https://onthebench.ai/gateways/higress.md) - [Kong](https://onthebench.ai/gateways/kong.md) - [LiteLLM · Python](https://onthebench.ai/gateways/litellm-python.md) - [LiteLLM · Rust](https://onthebench.ai/gateways/litellm-rust.md) - [One-API](https://onthebench.ai/gateways/one-api.md) - [Plano](https://onthebench.ai/gateways/plano.md) - [Portkey](https://onthebench.ai/gateways/portkey.md) - [TensorZero](https://onthebench.ai/gateways/tensorzero.md) ## Retrieval alternatives Markdown pages also have text/plain .txt twins. Use the HTML URL without .md if your tool rejects text/markdown. - [Compact CSV](https://onthebench.ai/gateways/data.csv) - [Schema index](https://onthebench.ai/api/v1/schema) ## Example workflow Compare Busbar and Bifrost: read the gateway catalog, then [compare OpenAI to OpenAI p99 added latency](https://onthebench.ai/api/v1/compare?a=busbar&b=bifrost&ingress=openai&egress=openai&metric=added_latency_p99_us). Inspect comparability.status before reporting a ratio; read measured_at, definitions, and provenance.run_url. Cite both runs and methodology with the numbers and caveats. ## Agent guidance - The dashboard uses representative cells that can differ between gateways; do not rank those rows as like-for-like. - Choose explicit ingress and egress (upstream) protocols in /api/v1/measurements and /api/v1/compare. - Measurement pages are capped at 1 MiB of UTF-8 JSON. Always follow next until null, even when fewer than limit rows are returned. Use gateway, ingress, egress and metric filters to narrow retrieval. - Inspect comparability.status: only comparable supports a ratio. Missing evidence is not proof of comparability. - Distinguish measured_at from generated_at and published_at (unknown when null). Age is relative to age_as_of, not retrieval time. current_metadata contains dated repository statistics, not run-time statistics. - Never treat null, unavailable, or below-resolution values as zero. - Cite provenance.run_url with gateway_build, measured_at, the protocol pair, metric definition, and methodology. Verify raw_result_url against raw_sha256 when checking evidence. - [Measurement JSON Schema](https://onthebench.ai/api/v1/schema/measurement.json) ## Interpretation 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.