OpenVLM MMT-Bench — leaderboard
OpenCompass OpenVLM evaluation of MMT-Bench: multimodal tasks covering visual recognition, localization, OCR, counting, relation reasoning, visual math, and hallucination detection.
Metric: Accuracy (%). Source: huggingface.co. Status: saturation imminent. 207 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | InternVL3-78B | 72.6 |
| 2 | InternVL3-38B | 71.9 |
| 3 | Qwen 2 VL 72B | 71.2 |
| 4 | GPT-4.1 (2025-04-14) | 71.1 |
| 5 | InternVL3-14B | 69.7 |
| 6 | Grok 2 (1212) | 67.6 |
| 7 | Ovis2-8B | 66.6 |
| 8 | InternVL3-8B | 64.9 |
| 9 | GPT-4.1 Mini | 63.4 |
| 10 | Qwen 2 VL 7B | 63.4 |
| 11 | Gemini 1.5 Pro | 63.4 |
| 12 | Gemini 1.5 Flash | 62.6 |
| 13 | Claude 3.5 Sonnet | 61.8 |
| 14 | MiniCPM-V-2.6 | 60.8 |
| 15 | Gemma 3 27B | 60.8 |
Interactive version: theaggregate.ai/benchmark?slug=openvlm-mmt-bench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.