Skip to content

Benchmark

MMBench V1.1 EN

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

Models measured
50
Scores recorded
50
Sources
1
IQ weight
1.06

Score distribution

76.3%
Mean
79.4%
Median
13.2%
Std dev
64.4%
10th percentile
86.9%
90th percentile
1.2%
Min
1.2%88.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.06 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
1Gemini 2.5 Pro· OpenVLM Leaderboard data88.5%
2SenseNova-V6-Pro· OpenVLM Leaderboard data88.4%
3Qwen2.5 VL 72B Instruct· OpenVLM Leaderboard data88.3%
4InternVL3 78B· OpenVLM Leaderboard data87.5%
5internvl3-38b· OpenVLM Leaderboard data87%
6GPT 4.1· OpenVLM Leaderboard data86.9%
7ChatGPT-4o Latest· OpenVLM Leaderboard data86.8%
8GPT 5· OpenVLM Leaderboard data86.8%
9GPT-5 mini· OpenVLM Leaderboard data86.2%
10Qwen2-VL-72B· OpenVLM Leaderboard data86.1%
11gemini-1.5-pro-002· OpenVLM Leaderboard data84.4%
12Hunyuan Standard Vision· OpenVLM Leaderboard data83.9%
13GPT-4.5 Preview· OpenVLM Leaderboard data83.5%
14Grok 2 Vision 1212· OpenVLM Leaderboard data83.3%
15GPT-4.1 mini· OpenVLM Leaderboard data83.2%
16Qwen2.5-VL 7B Instruct· OpenVLM Leaderboard data82.4%
17DeepSeek VL2· OpenVLM Leaderboard data82.3%
18internvl3-8b· OpenVLM Leaderboard data82.3%
19gemini-1.5-flash-002· OpenVLM Leaderboard data81.6%
20InternVL2 26B· OpenVLM Leaderboard data81.5%
21GLM 4 Vision Plus· OpenVLM Leaderboard data79.7%
22Gemma 3 27B IT· OpenVLM Leaderboard data79.6%
23Llama 3.2 90B Vision Instruct· OpenVLM Leaderboard data79.5%
24Phi 4 Multimodal Instruct· OpenVLM Leaderboard data79.5%
25Molmo 72B· OpenVLM Leaderboard data79.4%
26GLM 4 Vision· OpenVLM Leaderboard data79.2%
27MiniCPM-V 2.6· OpenVLM Leaderboard data79%
28GPT-5 nano· OpenVLM Leaderboard data78.9%
29Gemini 1.5 Flash· OpenVLM Leaderboard data77.1%
30qwen2.5-vl-3b-instruct· OpenVLM Leaderboard data77%
31Moonshot V1 8K· OpenVLM Leaderboard data76.8%
32Yi Vision· OpenVLM Leaderboard data76.6%
33InternVL2 4B· OpenVLM Leaderboard data75.8%
34Qwen-VL Max· OpenVLM Leaderboard data75.4%
35Pixtral 12B· OpenVLM Leaderboard data75.3%
36Gemma 3 12B IT· OpenVLM Leaderboard data74.9%
37Gemini 1.5 Pro· OpenVLM Leaderboard data74.6%
38Phi-3.5 Vision Instruct· OpenVLM Leaderboard data72.1%
39Gemini 2.0 Flash· OpenVLM Leaderboard data70.4%
40Molmo 7B-D· OpenVLM Leaderboard data70.3%
41gemini-1.0-pro· OpenVLM Leaderboard data70.2%
42Moondream2· OpenVLM Leaderboard data67.7%
43Llama 3.2 11B Vision Instruct· OpenVLM Leaderboard data67.5%
44Gemma 3 4B· OpenVLM Leaderboard data66.1%
45GPT-4.1 nano· OpenVLM Leaderboard data64.4%
46Qwen-VL Plus· OpenVLM Leaderboard data64.3%
47Claude 3 Sonnet· OpenVLM Leaderboard data64.2%
48Claude 3 Opus· OpenVLM Leaderboard data60.1%
49Claude 3 Haiku· OpenVLM Leaderboard data58%
50kosmos-2· OpenVLM Leaderboard data1.2%

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