Bring your own eval

Your runs. Your models. Your cost per solved task.

Drop a CSV. The browser computes pass rate, a 95% Wilson interval, and cost per solved task. Nothing leaves this page. The CLI is the authoritative implementation; this page is a low-friction preview for design partners.

Columns: task,model,passed,cost,attempts,quality. passed accepts true/false/1/0/yes/no. Empty cost is missing, not zero.

Or run the real CLI

npm install modelpick
modelpick eval evals.csv --task coding-agent --html report.html

The CLI checks every model against the catalog, validates booleans and numbers, and emits a self-contained report. This page intentionally does not call an API.