Why we open-sourced our eval harness
Sam, Engineering·6/14/2026·5 min read
Evals are how we know a model actually works. We're opening ours.
Evals are how we know a model actually works. We're opening ours.
Our biggest release ever — new models, new UI, and 10x faster signals.
EngineeringThe architecture, tooling and evals behind our conversational quant.
PerspectiveWhere we are, where we're going, and what actually works in production.