Model comparison

Benchmarks

Settings-extraction agent graded against a mock ground truth: correctness, wasted steps, latency, tokens, and cost per model.

No benchmarks recorded yet. Run benchmark --mock <name> --models a,b from the CLI.