aiaimodel.cards
← All benchmarks

MM MTBench

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. 1Ministral 3 14B · Mistral AIbest value
    8.49 Score
  2. 2Ministral 3 8B · Mistral AIbest value
    8.08 Score
  3. 3Ministral 3 3B · Mistral AIbest value
    7.83 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

  • Ministral 3 14B model card, Benchmark Results

    Mistral's tables give fractions (0.551); shown here as percentages. Qwen3 14B is in non-thinking mode for the Instruct rows and thinking mode for the Reasoning rows; it is not multimodal (no MM MTBench). Gemma3-12B appears only in the Instruct table. The model card also has a Base-model table, not shown.

  • Ministral 3 8B model card, Benchmark Results

    Mistral's tables give fractions (0.551); shown here as percentages. Qwen3-VL-8B is the Instruct model in the Instruct rows and the Thinking model in the Reasoning rows. The model card also has a Base-model table, not shown.

  • Ministral 3 3B model card, Benchmark Results

    Mistral's tables give fractions (0.551); shown here as percentages. Qwen3-VL-4B is the Instruct model in the Instruct rows and the Thinking model in the Reasoning rows; Qwen3-VL-2B and Gemma3-4B appear only in the Instruct table. The model card also has a Base-model table, not shown.

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