Skip to content

Benchmark

MMT-Bench (val)

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

Models measured
34
Scores recorded
34
Sources
1
IQ weight
1.23

Score distribution

57.6%
Mean
60.8%
Median
14.7%
Std dev
50.4%
10th percentile
71.1%
90th percentile
7.2%
Min
7.2%72.6%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.23 for chat models, measured over 34 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 data72.6%
2internvl3-38b· OpenVLM Leaderboard data71.9%
3Qwen2-VL-72B· OpenVLM Leaderboard data71.2%
4GPT 4.1· OpenVLM Leaderboard data71.1%
5Qwen2.5 VL 72B Instruct· OpenVLM Leaderboard data69%
6Grok 2 Vision 1212· OpenVLM Leaderboard data67.6%
7Hunyuan Standard Vision· OpenVLM Leaderboard data66.1%
8internvl3-8b· OpenVLM Leaderboard data64.9%
9GLM 4 Vision Plus· OpenVLM Leaderboard data64%
10Gemini 1.5 Pro· OpenVLM Leaderboard data63.4%
11GPT-4.1 mini· OpenVLM Leaderboard data63.4%
12Qwen-VL Plus· OpenVLM Leaderboard data63.1%
13Yi Vision· OpenVLM Leaderboard data62.9%
14Gemini 1.5 Flash· OpenVLM Leaderboard data62.6%
15Qwen2.5-VL 7B Instruct· OpenVLM Leaderboard data62.1%
16qwen2.5-vl-3b-instruct· OpenVLM Leaderboard data61%
17Gemma 3 27B IT· OpenVLM Leaderboard data60.8%
18MiniCPM-V 2.6· OpenVLM Leaderboard data60.8%
19InternVL2 26B· OpenVLM Leaderboard data60.4%
20Llama 3.2 90B Vision Instruct· OpenVLM Leaderboard data60.4%
21Phi 4 Multimodal Instruct· OpenVLM Leaderboard data60.4%
22Gemma 3 12B IT· OpenVLM Leaderboard data59.7%
23Pixtral 12B· OpenVLM Leaderboard data59.4%
24Qwen-VL Max· OpenVLM Leaderboard data58.3%
25Molmo 7B-D· OpenVLM Leaderboard data56.3%
26InternVL2 4B· OpenVLM Leaderboard data55.9%
27GPT-4.1 nano· OpenVLM Leaderboard data55.3%
28Gemma 3 4B· OpenVLM Leaderboard data53.9%
29Phi-3.5 Vision Instruct· OpenVLM Leaderboard data53.6%
30Llama 3.2 11B Vision Instruct· OpenVLM Leaderboard data53.1%
31Claude 3 Haiku· OpenVLM Leaderboard data50.4%
32kosmos-2· OpenVLM Leaderboard data25.5%
33Claude 3 Sonnet· OpenVLM Leaderboard data10.2%
34Moonshot V1 8K· OpenVLM Leaderboard data7.2%

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