AI Gateway Benchmarks
The same gateway-latency harness run against four target models. Some gateways don't support every target model so each family shows the participants that actually route to that model, scored against a direct-to-provider baseline.
Anthropic
Each gateway in the roster routed to Claude Haiku 4.5, measured against a direct-to-Anthropic baseline.
10 Gateways
Runs Weekly
View benchmarks
OpenAI
The same gateways routed to GPT-5.4 Mini instead, measured against a direct-to-OpenAI baseline.
9 Gateways
Runs Weekly
View benchmarks
Gemini
Each gateway in the roster routed to Gemini 3.6 Flash, measured against a direct-to-Gemini baseline.
9 Gateways
Runs Weekly
View benchmarks
Kimi
Each gateway in the roster routed to Kimi K3, measured against a direct-to-Moonshot baseline.
10 Gateways
Runs Weekly
View benchmarks