Skip to content

Benchmark

ChartQA

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

Models measured
32
Scores recorded
34
Sources
3
IQ weight
0.80

Score distribution

69.4%
Mean
83%
Median
26.2%
Std dev
30.3%
10th percentile
88.8%
90th percentile
0%
Min
0%90.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.80 for chat models, measured over 31 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude 3.5 Sonnet· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim90.8%
2Llama 4 Maverick· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim90%
3Qwen2.5 VL 72B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim89.5%
4Nova Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim89.2%
5Llama 4 Scout· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim88.8%
6Llama 4 Scoutinstruction-tuned table· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.88.8%
7Qwen2-VL-72B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
88.3%
8Pixtral Large· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim88.1%
9Mistral Small 3.2 24B Instruct2506· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim87.4%
10Qwen2.5-VL 7B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim87.3%
11Nova Lite· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim86.8%
12DeepSeek VL2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim86%
13internvl3-8b· OpenVLM Leaderboard data85.8%
14GPT 4o· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim85.7%
15Qwen2.5-Omni 7B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim85.3%
16InternVL2 26B· OpenVLM Leaderboard data85.1%
17Llama 3.2 11B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim83.4%
18Phi 4 Multimodal Instruct· OpenVLM Leaderboard data
Values from 2 sources
83%
19MiniCPM-V 2.6· OpenVLM Leaderboard data82.4%
20Phi-3.5 Vision Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim81.8%
21Pixtral 12B2409· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim81.8%
22InternVL2 4B· OpenVLM Leaderboard data80.7%
23Gemma 3 27B IT· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
78%
24Gemma 3 12B IT· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
75.7%
25Pixtral 12B· OpenVLM Leaderboard data72.2%
26Gemma 3 4B IT· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim68.8%
27Molmo 7B-D· OpenVLM Leaderboard data43.7%
28Gemma 3 4B· OpenVLM Leaderboard data33.7%
29GPT-4.1 mini· OpenVLM Leaderboard data33.7%
30GPT 4.1· OpenVLM Leaderboard data33.3%
31GPT-4.1 nano· OpenVLM Leaderboard data30.3%
32Gemini 1.5 Flash· OpenVLM Leaderboard data29.7%
33kosmos-2· OpenVLM Leaderboard data5.9%
34Llama 3.2 11B Vision Instruct· OpenVLM Leaderboard data0%

Data credits: Data by LLM Stats · OpenVLM Leaderboard (OpenCompass / VLMEvalKit), Apache-2.0 (licence) · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.