Skip to content

Benchmark

MTVQA

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

Models measured
30
Scores recorded
30
Sources
2
IQ weight
1.03

Score distribution

25.4%
Mean
25.8%
Median
8.8%
Std dev
15.3%
10th percentile
32.5%
90th percentile
2%
Min
2%50.6%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.03 for chat models, measured over 30 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Seed 2.0 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim50.6%
2GPT-4.1 mini· OpenVLM Leaderboard data36.8%
3GPT 4.1· OpenVLM Leaderboard data34.1%
4InternVL3 78B· OpenVLM Leaderboard data32.5%
5internvl3-38b· OpenVLM Leaderboard data32.4%
6Gemini 1.5 Pro· OpenVLM Leaderboard data32%
7Qwen2.5 VL 72B Instruct· OpenVLM Leaderboard data31.5%
8Qwen2-VL-72B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
30.9%
9Gemini 1.5 Flash· OpenVLM Leaderboard data30.5%
10internvl3-8b· OpenVLM Leaderboard data30.4%
11Qwen-VL Max· OpenVLM Leaderboard data29.2%
12Qwen2.5-VL 7B Instruct· OpenVLM Leaderboard data29%
13Gemma 3 27B IT· OpenVLM Leaderboard data28.9%
14GPT-4.1 nano· OpenVLM Leaderboard data27.6%
15Qwen-VL Plus· OpenVLM Leaderboard data25.8%
16GLM 4 Vision Plus· OpenVLM Leaderboard data25.6%
17Gemma 3 12B IT· OpenVLM Leaderboard data23.8%
18qwen2.5-vl-3b-instruct· OpenVLM Leaderboard data23.8%
19Hunyuan Standard Vision· OpenVLM Leaderboard data23.4%
20Grok 2 Vision 1212· OpenVLM Leaderboard data22.6%
21Gemma 3 4B· OpenVLM Leaderboard data22.1%
22Pixtral 12B· OpenVLM Leaderboard data20.2%
23Moonshot V1 8K· OpenVLM Leaderboard data20%
24InternVL2 26B· OpenVLM Leaderboard data17.7%
25MiniCPM-V 2.6· OpenVLM Leaderboard data17%
26InternVL2 4B· OpenVLM Leaderboard data15.5%
27Llama 3.2 11B Vision Instruct· OpenVLM Leaderboard data15.3%
28Phi 4 Multimodal Instruct· OpenVLM Leaderboard data14.3%
29Molmo 7B-D· OpenVLM Leaderboard data13.1%
30kosmos-2· OpenVLM Leaderboard data2%

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