Benchmarks

Add a benchmark with an assistant

Open Benchmarks → Add. The shared dialog preselects Benchmark; enter its name and source, then copy the prompt to your coding assistant. It uses /import-benchmark. See the shared import procedure.

Accept when

A recorded pin, approved glue, reported control results, a small trial with inspectable jobs, stated gaps.

Trials and control make paid model calls

A cheap model; each trial's tokens and cost are reported.

• Supported benchmarks

• Add manually