Get updates.

Create a free account. You will get a short editorial note when a state-of-the-art score changes on a benchmark you follow — and nothing else. No newsletter, no drip sequence, no funnel.

The registry itself is, and will remain, free to read without an account.

§ 02

What an account is for.

Follow a benchmark

Star any benchmark and receive a short note when the state of the art changes — with a link to the reproduction.

Submit a result

Post a checkpoint and a reproduction script. Verified runs are published with your name on the annotation.

Save comparisons

Keep a list of models you are evaluating across modalities; export to CSV or JSON from the dashboard.

Query the API

Signed requests raise the anonymous rate limit from 1,000 to 10,000 calls per hour. Read endpoints remain open.

Already have an account? Sign in to pick up where you left off. Prefer not to create one yet? The registry is fully readable without one.