MM-MT-Bench: leaderboard
92 image-grounded conversations (69 single-turn, 23 multi-turn) over charts, tables, PDF pages and diagrams from Mistral AI (2024); a GPT-4o judge rates correctness and completeness from 1 to 10.
Metric: Score (points). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet | 7.5 | #350 |
| 2 | Qwen 2 VL 72B | 6.59 | #460 |
| 3 | Pixtral-12B | 6.05 | #808 |
| 4 | Llama 3.2 90B | 5.5 | #596 |
| 5 | Claude 3 Haiku | 5.46 | #803 |
| 6 | Qwen 2 VL 7B | 5.45 | #750 |
| 7 | Llama 3.2 11B | 4.79 | #1199 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mm-mt-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.