MMBench V1.1: leaderboard
2,974 bilingual (English/Chinese) multiple-choice questions evaluating vision-language models across 20 ability dimensions spanning perception and reasoning capabilities.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 284 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 88.3 | #152 |
| 2 | InternVL3-78B | 87.7 | #352 |
| 3 | InternVL2.5-78B | 87.55 | #431 |
| 4 | InternVL3-38B | 86.8 | #405 |
| 5 | GPT-5 | 86.6 | #98 |
| 6 | GPT-4.1 (2025-04-14) | 86.55 | #250 |
| 7 | GPT-5 Mini (2025-08-07) | 86.15 | #172 |
| 8 | GPT-4o ChatGPT | 86 | |
| 9 | Qwen 2 VL 72B | 85.95 | #460 |
| 10 | InternVL3-14B | 83.6 | #505 |
| 11 | GPT-4.5 | 83.4 | #274 |
| 12 | Grok 2 (1212) | 82.9 | #496 |
| 13 | Gemini 1.5 Pro (002) | 82.75 | #349 |
| 14 | Qwen 2.5 VL 7B | 82.25 | #643 |
| 15 | InternVL3-8B | 82.1 | #620 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmbench-v1-1 · How It Works · Data refreshed daily, snapshot 2026-10-20.