One API for thousands of benchmarks.One API for thousands of benchmarks.

Run the install command at the top of this page, or give the setup prompt to your coding agent. The CLI signs you in through your browser and shows the benchmarks and models you can run. New accounts get $5 of credit.

You can. EvalRouter removes the work around them: no sandboxes, provider keys, retries or result storage to manage. Hundreds of tasks run in parallel, and every run is recorded so results stay comparable over time.

EvalRouter runs each benchmark with its authors' maintained implementation and keeps their evaluation method. Any change to the method ships as a new, labeled version, so a score always says exactly what it measured.

Every benchmark in the catalog, on managed models charged at provider cost or on your own endpoint. Agents run on your own machine with evalrouter run --agent. EvalRouter checks that a benchmark and model can run together before you spend anything.

Before a job starts, its preview shows an estimated cost wherever recorded usage supports one. Run a small sample first to see real numbers.

You pay what models and sandboxes cost, plus 12.5%. There are no minimums. For example, $10 of compute and inference costs $11.25.

Private benchmarks, datasets, runs and traces stay in your workspace. Uploaded data is never executed, and sandboxes run with no credentials and no outbound network.

Volume discounts, invoicing, higher limits, single sign-on, direct support, Integrity Check and custom benchmarks and environments built with our team. Email info@kimpton.ai.