Skip to content

Benchmark

CCBench

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

Models measured
39
Scores recorded
39
Sources
1
IQ weight
0.68

Score distribution

51.9%
Mean
55.5%
Median
19.1%
Std dev
27.6%
10th percentile
73.9%
90th percentile
1.4%
Min
1.4%83.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.68 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 data83.5%
2internvl3-38b· OpenVLM Leaderboard data80.6%
3internvl3-8b· OpenVLM Leaderboard data77.8%
4InternVL2 26B· OpenVLM Leaderboard data73.9%
5Qwen2.5 VL 72B Instruct· OpenVLM Leaderboard data73.9%
6GPT 4.1· OpenVLM Leaderboard data73.5%
7Hunyuan Standard Vision· OpenVLM Leaderboard data72%
8Qwen2-VL-72B· OpenVLM Leaderboard data69.8%
9InternVL2 4B· OpenVLM Leaderboard data66.3%
10GPT-4.1 mini· OpenVLM Leaderboard data64.3%
11GLM 4 Vision· OpenVLM Leaderboard data63.9%
12Qwen-VL Max· OpenVLM Leaderboard data63.5%
13GLM 4 Vision Plus· OpenVLM Leaderboard data62.5%
14qwen2.5-vl-3b-instruct· OpenVLM Leaderboard data62%
15Moonshot V1 8K· OpenVLM Leaderboard data60.2%
16Qwen2.5-VL 7B Instruct· OpenVLM Leaderboard data59.2%
17Grok 2 Vision 1212· OpenVLM Leaderboard data57.3%
18DeepSeek VL2· OpenVLM Leaderboard data55.7%
19MiniCPM-V 2.6· OpenVLM Leaderboard data55.7%
20Phi 4 Multimodal Instruct· OpenVLM Leaderboard data55.5%
21Qwen-VL Plus· OpenVLM Leaderboard data55.1%
22Llama 3.2 90B Vision Instruct· OpenVLM Leaderboard data54.1%
23Gemma 3 27B IT· OpenVLM Leaderboard data53.3%
24gemini-1.0-pro· OpenVLM Leaderboard data52.5%
25Gemini 1.5 Flash· OpenVLM Leaderboard data52.4%
26Yi Vision· OpenVLM Leaderboard data50.6%
27Gemma 3 12B IT· OpenVLM Leaderboard data49%
28Pixtral 12B· OpenVLM Leaderboard data37.6%
29Gemma 3 4B· OpenVLM Leaderboard data36.9%
30Molmo 7B-D· OpenVLM Leaderboard data36.7%
31GPT-4.1 nano· OpenVLM Leaderboard data31.2%
32Llama 3.2 11B Vision Instruct· OpenVLM Leaderboard data30.2%
33Gemini 1.5 Pro· OpenVLM Leaderboard data28.4%
34Claude 3 Sonnet· OpenVLM Leaderboard data27.8%
35Phi-3.5 Vision Instruct· OpenVLM Leaderboard data27.6%
36Claude 3 Opus· OpenVLM Leaderboard data26.3%
37Claude 3 Haiku· OpenVLM Leaderboard data24.5%
38Moondream2· OpenVLM Leaderboard data15.5%
39kosmos-2· OpenVLM Leaderboard data1.4%

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