tacit

What changed

Every leaderboard release, newest first. A change is a new model, a new version of a model, a re-run, a new #1 on a board, or a 5% target newly held or missed. A rank that moves only because another model was added is not listed.

Follow it with the RSS feed. Model authors can embed a badge that updates with each release.

  1. Leaderboard v0.2

    3 models added (DREX Max, Elephant-1, Liquid d1.1); DREX Max new #1 (all languages pooled, Polish, English, German)

    • New #1, all languages pooled: DREX Max, 99.0% automated at 5% (previously Mammoth-1, 98.6%).
    • New #1, Polish: DREX Max, 99.2% automated at 5% (previously Mammoth-1, 98.8%).
    • New #1, English: DREX Max, 98.7% automated at 5% (previously basal-omni-1.0-ultra (basal 2.3.3), 98.1%).
    • New #1, German: DREX Max, 99.5% automated at 5% (previously basal-omni-1.0-ultra (basal 2.3.3), 98.7%).
    • DREX Max (Nace.AI) added at #1 of 20 (all languages pooled), 99.0% automated at 5% (Polish #1, English #1, German #1, Norwegian Bokmål #2).
    • Elephant-1 (Baseten) added at #9 of 20 (all languages pooled), 93.9% automated at 5% (Polish #7, English #9, German #10, Norwegian Bokmål #9).
    • Liquid d1.1 (Liquid AI) added at #10 of 20 (all languages pooled), 92.9% automated at 5% (Polish #10, English #12, German #9, Norwegian Bokmål #10).
    • Norwegian Bokmål: Elephant-1 misses the 5% target, 5.8% realised error (41 of 703 automated).

    Leaderboard v0.2

  2. Leaderboard v0.2

    Tacit v0.2 released, 17 models

    • Tacit v0.2 released: Polish, English, German, Norwegian Bokmål, 17 models. Highest share automated at a 5% error target (all languages pooled): Mammoth-1, 98.6%.

    Leaderboard v0.2

  3. Leaderboard v0.1

    10 models added; Liquid d1-ultra new #1 (Polish)

    • New #1, Polish: Liquid d1-ultra, 94.2% automated at 5% (previously Jev 1.13, 89.7%).
    • Jev 1.14 added, successor of Jev 1.13: 89.7% → 92.0% automated at 5%, #2 of 16.
    • Liquid d1-ultra (Liquid AI) added at #1 of 16, 94.2% automated at 5%.
    • basal-omni-1.0-ultra (basal) added at #3 of 16, 91.6% automated at 5%.
    • pplx-decider-v1.2 (Perplexity) added at #4 of 16, 90.6% automated at 5%.
    • Mercury Decide (Inception) added at #4 of 16, 90.6% automated at 5%.
    • Microsoft-Decision-1 (Microsoft) added at #7 of 16, 89.1% automated at 5%.
    • basal-omni-1.1-max (basal) added at #8 of 16, 87.1% automated at 5%.
    • Cloudflare Clef (Cloudflare) added at #11 of 16, 84.4% automated at 5%.
    • Respan Span-02 (Respan) added at #12 of 16, 84.2% automated at 5%.
    • GPT-6 Luna Decisions (OpenAI) added at #16 of 16, 47.5% automated at 5%.

    Frozen leaderboard v0.1

  4. Leaderboard v0.1

    Tacit v0.1 released, 6 models

    • Tacit v0.1 released: Polish, 6 models. Highest share automated at a 5% error target: Jev 1.13, 89.7%.

    Frozen leaderboard v0.1