Benchmarks / AI Inference
AI Inference benchmarks
Time to first token, streaming throughput, cold starts and tail latency across AI inference providers — measured continuously, not quoted from launch blogs.
Planned — not yet measuringProviders on the roadmap
OpenAIAnthropicGoogleGroqTogether AIFireworks AI
Same rules as compute: deterministic workloads, open data, no affiliations. The test suite will ship in the same CLI — providerbench run -t ttft — so anyone can verify from their own network.
Want this category sooner?
The benchmark framework is one Go interface — category suites are designed in the open. Propose metrics or contribute tests.