MIBench (Multimodal Interaction): leaderboard
Metric: Overall accuracy (%) on the full MIBench (multiple-choice image-text tasks; for vision- and text-centric samples the question is posed under several context variations and per-sample accuracy averages them); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 235B A22B Instruct | 67.89 | #264 |
| 2 | Qwen 2.5 VL 72B Instruct | 63.87 | #364 |
| 3 | Qwen 2.5 VL 32B Instruct | 61.22 | #443 |
| 4 | Qwen 3 VL 8B Instruct | 60.85 | #401 |
| 5 | InternVL3-8B | 58.82 | #606 |
| 6 | Qwen 2 VL 7B Instruct | 56.09 | #816 |
| 7 | Qwen 2.5 VL 7B Instruct | 54.56 | #643 |
| 8 | InternVL2-8B | 54.23 | #826 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=mibench-multimodal-interaction · How It Works · Data refreshed daily, snapshot 2026-10-11.