MMBench V1.1: leaderboard
Metric: Accuracy (%). Source: huggingface.co. 284 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 88.3 |
| 2 | InternVL3-78B | 87.7 |
| 3 | InternVL2.5-78B | 87.55 |
| 4 | InternVL3-38B | 86.8 |
| 5 | GPT-5 | 86.6 |
| 6 | GPT-4.1 (2025-04-14) | 86.55 |
| 7 | GPT-5 Mini (2025-08-07) | 86.15 |
| 8 | GPT-4o ChatGPT | 86 |
| 9 | Qwen 2 VL 72B | 85.95 |
| 10 | GPT-4.5 | 83.4 |
| 11 | Grok 2 (1212) | 82.9 |
| 12 | Gemini 1.5 Pro (002) | 82.75 |
| 13 | InternVL3-8B | 82.1 |
| 14 | Claude 3.5 Sonnet (20241022) | 81.7 |
| 15 | Gemini 1.5 Flash (002) | 81.3 |
Interactive version: theaggregate.ai/benchmark?slug=mmbench-v1-1 · How It Works · Data refreshed daily, snapshot 2026-09-05.