ChartQA (test)

Name: ChartQA (test)
Published: 2022-03-19

Multimodal · 2022-03-19

ChartQA asks visual and logical reasoning questions about real-world charts. This row is the 2,500-question test split, its human-written and machine-generated halves combined, scored by relaxed accuracy, which accepts numeric answers within 5% of the reference and requires an exact match otherwise. ChartQAPro is a separate benchmark. Higher is better.

Top models (higher is better)

Model	Score
LFM3-VL-2B	92.7
Claude 3.5 Sonnet (June 2024)	90.8
Claude 3.5 Sonnet (Oct 2024)	90.8
Qwen3 VL 235B A22B Thinking	90.3
Qwen3-VL-235B-A22B-Instruct	90.3
GPT-5.6 Luna	90.2
Nemotron Nano 12B V2 VL BF16	89.8
Qwen2.5-VL-72B-Instruct	89.5
Qwen3 Omni 30B A3B Thinking	89.5
Gemma 4 12B IT	89.4
Qwen3 VL 30B A3B Thinking	89.4
Qwen3 VL 32B Thinking	89.0
Llama 4 Scout Instruct	88.8
Qwen3 VL 4B Thinking	88.8
Qwen3 VL 8B Thinking	88.6
Qwen3 VL 32B Instruct	88.5
InternVL2.5-78B	88.3
Qwen2 VL 72B Instruct	88.3
Gemini 3.5 Flash-Lite	88.2
Pixtral Large	88.1
InternVL3.5-241B-A28B	88.0
Mistral Small 3.2 24B Instruct 2506	87.4
Qwen2.5-VL-7B-Instruct	87.3
Gemini 1.5 Pro (May 2024)	87.2
Haiku 3.5	87.2
Qwen3 Omni 30B A3B Instruct	86.8
Qwen3 VL 30B A3B Instruct	86.8
Qwen3 VL 2B Thinking	86.6
Opus 4.1	86.2
NVLM-D-72B	86.0
Qwen3.5-4B	85.8
GPT-4o (2024-05-13)	85.7
Gemini 1.5 Flash (May 2024 report)	85.4
Llama 4 Maverick	85.3
Qwen3.5-2B	85.2
GPT-4o (2024-08-06)	85.2
Qwen3 VL 4B Instruct	84.6
Qwen2.5-VL-3B-Instruct	84.0
Gemini 1.5 Pro 002	83.8
Llama 3.2 11B Vision Instruct	83.4