Skip to content

Benchmark

BLINK

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

Models measured
47
Scores recorded
49
Sources
3
IQ weight
0.74

Score distribution

56.4%
Mean
55.8%
Median
11.2%
Std dev
42.1%
10th percentile
69.1%
90th percentile
34.4%
Min
34.4%81.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.74 for chat models, measured over 47 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Doubao Seed 2.1 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim81.4%
2Doubao Seed 2.1 Turbo· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim79.4%
3Doubao Seed 1.8· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim74.3%
4Seed 2.0 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim73.4%
5Qwen3 VL 235B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.70.7%
6Qwen3 VL 8B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.69.1%
7Qwen: Qwen3 VL 8B Thinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.68.7%
8GPT 4.1· OpenVLM Leaderboard data68.5%
9Qwen3 VL 32B Thinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.68.5%
10qwen3-vl-30b· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.67.7%
11Qwen3 VL 32B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.67.3%
12Qwen3 VL 235Bthinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.67.1%
13InternVL3 78B· OpenVLM Leaderboard data66.4%
14qwen3-vl-30bthinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.65.4%
15internvl3-38b· OpenVLM Leaderboard data64.2%
16Qwen2.5 VL 72B Instruct· OpenVLM Leaderboard data63.2%
17Phi 4 Multimodal Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
61.3%
18Qwen2-VL-72B· OpenVLM Leaderboard data60.5%
19GLM 4 Vision Plus· OpenVLM Leaderboard data59.5%
20Gemini 1.5 Pro· OpenVLM Leaderboard data59.1%
21GPT-4.1 mini· OpenVLM Leaderboard data59%
22Phi-3.5 Vision Instruct· OpenVLM Leaderboard data58.3%
23Gemini 1.5 Flash· OpenVLM Leaderboard data57.7%
24InternVL2 26B· OpenVLM Leaderboard data56.9%
25Yi Vision· OpenVLM Leaderboard data56.7%
26Moonshot V1 8K· OpenVLM Leaderboard data55.8%
27Hunyuan Standard Vision· OpenVLM Leaderboard data55.4%
28internvl3-8b· OpenVLM Leaderboard data55.4%
29Qwen2.5-VL 7B Instruct· OpenVLM Leaderboard data55.3%
30MiniCPM-V 2.6· OpenVLM Leaderboard data55.2%
31Grok 2 Vision 1212· OpenVLM Leaderboard data53.7%
32Gemma 3 27B IT· OpenVLM Leaderboard data53.5%
33Qwen-VL Plus· OpenVLM Leaderboard data52.9%
34Pixtral 12B· OpenVLM Leaderboard data50.4%
35Gemma 3 12B IT· OpenVLM Leaderboard data50.3%
36Molmo 7B-D· OpenVLM Leaderboard data49.7%
37qwen2.5-vl-3b-instruct· OpenVLM Leaderboard data49.1%
38Llama 3.2 90B Vision Instruct· OpenVLM Leaderboard data48%
39InternVL2 4B· OpenVLM Leaderboard data45.8%
40Gemini Pro· EvalAI data45.7%
41GPT-4.1 nano· OpenVLM Leaderboard data45.2%
42LLaVA 1.6 34B· EvalAI data45.1%
43Claude 3 Opus· EvalAI data44.1%
44Gemma 3 4B· OpenVLM Leaderboard data43.9%
45Qwen-VL Max· OpenVLM Leaderboard data
Values from 2 sources
43.6%
46Llama 3.2 11B Vision Instruct· OpenVLM Leaderboard data39.8%
47Claude 3 Sonnet· OpenVLM Leaderboard data39.3%
48Claude 3 Haiku· OpenVLM Leaderboard data37.5%
49kosmos-2· OpenVLM Leaderboard data34.4%

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