Skip to content

Benchmark

MM-Vet

Vision benchmarks measure how well models understand images. Scores are percentages.

Models measured
50
Scores recorded
50
Sources
2
IQ weight
0.87

Score distribution

64.4%
Mean
64.1%
Median
12.7%
Std dev
50.9%
10th percentile
79.9%
90th percentile
23.7%
Min
23.7%88.2%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.87 for chat models, measured over 50 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1SenseNova-V6-Pro· OpenVLM Leaderboard data88.2%
2Gemini 2.5 Pro· OpenVLM Leaderboard data83.3%
3internvl3-8b· OpenVLM Leaderboard data82.8%
4internvl3-38b· OpenVLM Leaderboard data81.1%
5InternVL3 78B· OpenVLM Leaderboard data80.7%
6Hunyuan Standard Vision· OpenVLM Leaderboard data79.9%
7GPT 4.1· OpenVLM Leaderboard data78.8%
8GPT 5· OpenVLM Leaderboard data77.6%
9ChatGPT-4o Latest· OpenVLM Leaderboard data76.9%
10Qwen2.5 VL 72B Instruct· OpenVLM Leaderboard data
Values from 2 sources
76.9%
11GPT-4.5 Preview· OpenVLM Leaderboard data75.3%
12gemini-1.5-pro-002· OpenVLM Leaderboard data74.6%
13GPT-5 mini· OpenVLM Leaderboard data74.6%
14GPT-4.1 mini· OpenVLM Leaderboard data74.3%
15Qwen2-VL-72B· OpenVLM Leaderboard data73.9%
16Gemini 2.0 Flash· OpenVLM Leaderboard data73.6%
17Gemma 3 27B IT· OpenVLM Leaderboard data71%
18gemini-1.5-flash-002· OpenVLM Leaderboard data69.7%
19Qwen2.5-VL 7B Instruct· OpenVLM Leaderboard data
Values from 2 sources
69.7%
20GLM 4 Vision Plus· OpenVLM Leaderboard data69.1%
21GPT-5 nano· OpenVLM Leaderboard data67.4%
22Grok 2 Vision 1212· OpenVLM Leaderboard data65.2%
23Gemma 3 12B IT· OpenVLM Leaderboard data64.9%
24Llama 3.2 90B Vision Instruct· OpenVLM Leaderboard data64.1%
25Gemini 1.5 Pro· OpenVLM Leaderboard data64%
26Gemini 1.5 Flash· OpenVLM Leaderboard data63.2%
27GPT-4.1 nano· OpenVLM Leaderboard data62.8%
28Qwen-VL Max· OpenVLM Leaderboard data61.8%
29Molmo 72B· OpenVLM Leaderboard data61.1%
30GLM 4 Vision· OpenVLM Leaderboard data60.7%
31DeepSeek VL2· OpenVLM Leaderboard data60%
32InternVL2 26B· OpenVLM Leaderboard data60%
33MiniCPM-V 2.6· OpenVLM Leaderboard data60%
34qwen2.5-vl-3b-instruct· OpenVLM Leaderboard data60%
35gemini-1.0-pro· OpenVLM Leaderboard data58.6%
36Pixtral 12B· OpenVLM Leaderboard data58.5%
37Gemma 3 4B· OpenVLM Leaderboard data57.8%
38Llama 3.2 11B Vision Instruct· OpenVLM Leaderboard data57.6%
39Yi Vision· OpenVLM Leaderboard data57.6%
40Qwen-VL Plus· OpenVLM Leaderboard data55.7%
41Moonshot V1 8K· OpenVLM Leaderboard data52%
42Phi 4 Multimodal Instruct· OpenVLM Leaderboard data51.9%
43Claude 3 Opus· OpenVLM Leaderboard data51.7%
44Claude 3 Sonnet· OpenVLM Leaderboard data51.7%
45InternVL2 4B· OpenVLM Leaderboard data50.9%
46Claude 3 Haiku· OpenVLM Leaderboard data46.4%
47Phi-3.5 Vision Instruct· OpenVLM Leaderboard data43.2%
48Molmo 7B-D· OpenVLM Leaderboard data41.5%
49Moondream2· OpenVLM Leaderboard data40.4%
50kosmos-2· OpenVLM Leaderboard data23.7%

Data credits: OpenVLM Leaderboard (OpenCompass / VLMEvalKit), Apache-2.0 (licence) · Data by LLM Stats