ModelCards
← All benchmarks

Object, person and logo image queries (3 tasks)

We haven't written a description for this benchmark yet; see the card notes below for how each maker ran it.

Your workloadRAG over docs · 12K in · 600 out · 50K requests/mo · 30% cached
  1. 1Marengo 3.5 · TwelveLabs
    63.5 score

“Best value”: no other model here reports a higher score at a lower cost for your workload. Scores are self-reported by each maker with different settings, so treat close calls as ties.

How each maker reported it

  • Marengo 3.5 technical launch post; launch post

    TwelveLabs' own evaluation. Default output dimensions: 512 for Marengo 3.5 and 3.0, 3,072 for Gemini Embedding 2 and Amazon Nova Multimodal Embeddings. Retrieval uses nDCG@10, composed retrieval R@5, classification and QA Hit@1; the production suites are TwelveLabs benchmark sets. ECE: lower is better. For MMEB-v2, TwelveLabs states only the improvement over Marengo 3.0 (text-to-video +6.29, video QA +15.26, image QA +15.81 points), not 3.0's own scores, so that column is blank.

Only each model's own maker's number is shown; scores a maker reports for rival models are left out.