EvalRouter
Run models and agents on benchmarks, then see what each one completed, what it cost and the evidence behind every score.
Pick a benchmark and a model in the catalog, check the quote's maximum charge, run it, then read the graded results and exports.
A run charges recorded usage against workspace credit, never more than its quote's cap. If it reaches the cap, its finished work stays available.
Start here#
- Run your first evaluation from the CLI.
- Use the Python SDK, the TypeScript SDK or the model API from code.
- Bring your own benchmark from a file or a Hugging Face dataset, where your deployment has private imports.
Model compatibility explains which combinations run. Quotes and billing explains holds and charges.