OpenVLM MM-Vet - Language Generation: leaderboard
MM-Vet Language Generation score from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): GPT-4-graded open-ended answers on the MM-Vet questions that require writing a longer, well-formed answer. A question can require several capabilities, so the capability subsets overlap. Score 0-100, higher is better.
Metric: Score. Source: huggingface.co. Saturation forecast: Around February 2027. 284 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 76.1 | #159 |
| 2 | InternVL3-8B | 75.8 | #624 |
| 3 | InternVL3-78B | 74.4 | #361 |
| 4 | InternVL3-38B | 73.5 | #413 |
| 5 | GPT-4.1 (2025-04-14) | 72.7 | #257 |
| 6 | GPT-4o ChatGPT | 70.9 | |
| 7 | InternVL3-14B | 69.6 | #506 |
| 8 | GPT-5 | 67.7 | #105 |
| 9 | GPT-4.1 Mini | 67.5 | #363 |
| 10 | GPT-5 Mini (2025-08-07) | 66.7 | #175 |
| 11 | GPT-4.5 | 65.2 | #278 |
| 12 | Gemini 1.5 Pro (002) | 64 | #368 |
| 13 | Qwen 2 VL 72B | 62.5 | #450 |
| 14 | GPT-4.1 Nano | 60.6 | #730 |
| 15 | Grok 2 (1212) | 60 | #502 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-mm-vet-language-generation · How It Works · Data refreshed daily, snapshot 2026-11-05.