Start here: what does your code need back?
Answer this one question first. Every API takes the same key, so switching later is a URL and payload change, but an eval run on the wrong one measures the wrong product.
Use cases
One line per job. Click a row for the full request body, cost and what to avoid. Start with these defaults and add parameters only when the use case needs them.
No match. Clear the filter.
Evals: match the benchmark to the endpoint
Run defaults first, then add one parameter at a time. Only compare against systems in the same latency class.
Latency classes:Web SearchContentssub-second retrieval·Answer~2.7s p50·ResearchFinance10s to 5 min
| Benchmark | Endpoint and config | Compare against |
|---|
Why evals go off the rails
Check these before you trust a bad result.
Endpoint quick reference Host, auth, price and default rate limit
| API | Endpoint | Auth | Price | Default limit |
|---|