ModelCards
← All benchmarks

Cohere Embed v4

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. 1voyage-4-large · Voyage AI
    8.2%
  2. 2voyage-multimodal-3.5 · Voyage AI
    3.5%
  3. 3voyage-multimodal-3.5 · Voyage AI
    2.3%

“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

  • voyage-4-large launch post (RTEB, 29 datasets)

    Voyage's own numbers: average NDCG@10 margin of voyage-4-large over each model across all 29 RTEB datasets. Per-dataset scores are only in charts and a linked spreadsheet. Voyage also says MoE gives accuracy comparable to a dense model with 4x the active parameters (≈75% lower inference cost at equal accuracy).

  • voyage-multimodal-3.5 launch post (NDCG@10; 18 multimodal + 38 text datasets)

    Voyage's own numbers: average NDCG@10 margin of voyage-multimodal-3.5 over each model. Visual documents: ViDoRe, ViDoRe v2, MIRACL-VISION; video: MSR-VTT, YouCook2, DiDeMo; text: 38 datasets in six domains. Voyage also reports +4.56% over Cohere Embed v4 across 15 visual document datasets. Full results in Voyage's linked spreadsheet.

  • voyage-multimodal-3.5 launch post (NDCG@10; 18 multimodal + 38 text datasets)

    Voyage's own numbers: average NDCG@10 margin of voyage-multimodal-3.5 over each model. Visual documents: ViDoRe, ViDoRe v2, MIRACL-VISION; video: MSR-VTT, YouCook2, DiDeMo; text: 38 datasets in six domains. Voyage also reports +4.56% over Cohere Embed v4 across 15 visual document datasets. Full results in Voyage's linked spreadsheet.

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