Skip to content

Benchmark

AI2D

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

Models measured
69
Scores recorded
71
Sources
2
IQ weight
1.11

Score distribution

82.4%
Mean
84.2%
Median
10%
Std dev
70.7%
10th percentile
92.7%
90th percentile
25.6%
Min
25.6%94.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.11 for chat models, measured over 68 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51Gemini 1.5 Pro· OpenVLM Leaderboard data79.1%
52InternVL2 4B· OpenVLM Leaderboard data79%
53Pixtral 12B· OpenVLM Leaderboard data79%
54Gemini 1.5 Flash· OpenVLM Leaderboard data78.5%
55Yi Vision· OpenVLM Leaderboard data78.2%
56Phi-3.5 Vision Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
78.1%
57Llama 3.2 11B Vision Instruct· OpenVLM Leaderboard data77.3%
58GLM 4 Vision· OpenVLM Leaderboard data76.1%
59GPT-4.1 mini· OpenVLM Leaderboard data76%
60Qwen-VL Max· OpenVLM Leaderboard data75.7%
61Gemma 3 4B IT· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim74.8%
62gemini-1.0-pro· OpenVLM Leaderboard data72.9%
63Gemma 3 4B· OpenVLM Leaderboard data70.7%
64Claude 3 Opus· OpenVLM Leaderboard data70.6%
65Claude 3 Sonnet· OpenVLM Leaderboard data69.9%
66Llama 3.2 90B Vision Instruct· OpenVLM Leaderboard data69.5%
67GPT-4.1 nano· OpenVLM Leaderboard data68%
68Qwen-VL Plus· OpenVLM Leaderboard data65.7%
69Claude 3 Haiku· OpenVLM Leaderboard data65.6%
70Moondream2· OpenVLM Leaderboard data58.8%
71kosmos-2· OpenVLM Leaderboard data25.6%

Showing 51–71 of 71 results.

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