Availability and limits

The initial public offering supports model evaluations through the EvalRouter CLI and model API: catalog discovery, model connections, quote review, bounded runs, status, cancellation, results and JSON/CSV/HTML exports.

The catalog and quote determine what your workspace can run. An installed command or a listed benchmark does not guarantee that every profile and model combination is available. Use exact active profile and model identifiers returned by discovery, then review compatibility and spending before submission.

Registration does not provide evaluation credit. Paid runs require sufficient available credit. The quoted platform cap is not a promise that an evaluation will fit; connected model providers may charge separately.

Keep the original quote and operation key after an uncertain submission. Partial, failed and cancelled results remain distinct from successful completion. A local timeout does not establish that server work has stopped.

Start with the CLI quickstart.

Use the model API reference for direct HTTP integrations.