Design benchmarks

No scores.Just results.

Same task. One shot. Open every result.

Latest benchmarks

    Compare models

    Benchmark task
    Let both models collaborate in AI Council

    Same task. One shot. No edits.

    Open each result and decide for yourself.

    Sametest
    Oneattempt
    Fullresult

    How testing works

    Every model receives the same fixed instruction for a test. We keep the first recorded result with no follow-up prompt or manual cleanup.

    The full result is the main point. You can open it, use it, and see what worked or failed.

    One run shows what the model made that time. It is not a general ranking of the model.