MMVet: leaderboard
218 open-ended questions over 200 images on recognition, OCR, knowledge, language generation, spatial awareness and math in 16 combinations, from NUS and Microsoft (2023); GPT-4 scores each answer.
Metric: Score. Source: huggingface.co. 284 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 83.3 |
| 2 | InternVL3-8B | 82.8 |
| 3 | InternVL3-38B | 81.1 |
| 4 | InternVL3-78B | 80.7 |
| 5 | Doubao-1.5-Pro | 79.8 |
| 6 | GPT-4.1 (2025-04-14) | 78.8 |
| 7 | GPT-5 | 77.6 |
| 8 | GPT-4o ChatGPT | 76.9 |
| 9 | GPT-4.5 | 75.3 |
| 10 | Gemini 1.5 Pro (002) | 74.6 |
| 11 | GPT-5 Mini (2025-08-07) | 74.6 |
| 12 | GPT-4.1 Mini | 74.3 |
| 13 | Qwen 2 VL 72B | 73.9 |
| 14 | Gemini 2.0 Flash | 73.6 |
| 15 | Gemma 3 27B | 71 |
Interactive version: theaggregate.ai/benchmark?slug=mmvet · How It Works · Data refreshed daily, snapshot 2026-09-05.