Skip to content

Benchmark

MMBench CN

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

Models measured
39
Scores recorded
39
Sources
1
IQ weight
1.19

Score distribution

74%
Mean
77.8%
Median
16.1%
Std dev
64.2%
10th percentile
86.7%
90th percentile
0.6%
Min
0.6%88.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.19 for chat models, measured over 39 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1InternVL3 78B· OpenVLM Leaderboard data88.8%
2Qwen2.5 VL 72B Instruct· OpenVLM Leaderboard data88.2%
3internvl3-38b· OpenVLM Leaderboard data86.9%
4GPT 4.1· OpenVLM Leaderboard data86.8%
5Qwen2-VL-72B· OpenVLM Leaderboard data86.7%
6GPT-4.1 mini· OpenVLM Leaderboard data85.5%
7Hunyuan Standard Vision· OpenVLM Leaderboard data85%
8Grok 2 Vision 1212· OpenVLM Leaderboard data84.8%
9Qwen2.5-VL 7B Instruct· OpenVLM Leaderboard data82.8%
10internvl3-8b· OpenVLM Leaderboard data82.7%
11DeepSeek VL2· OpenVLM Leaderboard data82.4%
12GLM 4 Vision Plus· OpenVLM Leaderboard data82.1%
13InternVL2 26B· OpenVLM Leaderboard data82.1%
14Gemma 3 27B IT· OpenVLM Leaderboard data80.7%
15GLM 4 Vision· OpenVLM Leaderboard data79.4%
16MiniCPM-V 2.6· OpenVLM Leaderboard data79.3%
17Moonshot V1 8K· OpenVLM Leaderboard data78.9%
18Gemini 1.5 Flash· OpenVLM Leaderboard data78.6%
19Yi Vision· OpenVLM Leaderboard data78.6%
20Llama 3.2 90B Vision Instruct· OpenVLM Leaderboard data77.8%
21Phi 4 Multimodal Instruct· OpenVLM Leaderboard data77.7%
22qwen2.5-vl-3b-instruct· OpenVLM Leaderboard data77.6%
23Gemma 3 12B IT· OpenVLM Leaderboard data77.1%
24Qwen-VL Max· OpenVLM Leaderboard data75.7%
25Molmo 7B-D· OpenVLM Leaderboard data75.2%
26Pixtral 12B· OpenVLM Leaderboard data74.4%
27gemini-1.0-pro· OpenVLM Leaderboard data74.3%
28InternVL2 4B· OpenVLM Leaderboard data73.9%
29Gemini 1.5 Pro· OpenVLM Leaderboard data73.8%
30Qwen-VL Plus· OpenVLM Leaderboard data70.7%
31Gemma 3 4B· OpenVLM Leaderboard data69.7%
32Phi-3.5 Vision Instruct· OpenVLM Leaderboard data66.1%
33Llama 3.2 11B Vision Instruct· OpenVLM Leaderboard data65.8%
34GPT-4.1 nano· OpenVLM Leaderboard data65.1%
35Claude 3 Sonnet· OpenVLM Leaderboard data64.2%
36Claude 3 Opus· OpenVLM Leaderboard data59.2%
37Claude 3 Haiku· OpenVLM Leaderboard data57.2%
38Moondream2· OpenVLM Leaderboard data28.7%
39kosmos-2· OpenVLM Leaderboard data0.6%

Data credits: OpenVLM Leaderboard (OpenCompass / VLMEvalKit), Apache-2.0 (licence)