Skip to content

Benchmark

Arena ELO Vision - Ocr

OCR benchmarks measure how well models read text in images and documents. Higher scores are better.

Models measured
108
Scores recorded
124
Sources
1
IQ weight
1.83

Score distribution

1,247.8
Mean
1,256.6
Median
59.1
Std dev
1,156.8
10th percentile
1,318.7
90th percentile
1,058.5
Min
1,058.51,336.4
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.83 for chat models, measured over 108 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
101GPT-4.1 mini2025-04-14· LMArena data1,187.5
102Gemini 2.5 Flash-Lite PreviewThinking, 06-17· LMArena data1,187.3
103Claude Sonnet 3.7Thinking 32K, 20250219· LMArena data1,180.4
104Claude Opus 420250514· LMArena data1,178.3
105Step-1o Turbo202506· LMArena data1,177.1
106Hunyuan Vision 1.5 Thinking· LMArena data1,175.1
107Mistral Medium 3.12508· LMArena data1,174.7
108Claude Sonnet 420250514· LMArena data1,173.1
109GLM 4.6v· LMArena data1,170.5
110Step3· LMArena data1,166.8
111Hunyuan Large Vision· LMArena data1,166.6
112Claude Sonnet 3.720250219· LMArena data1,156.8
113Gemma 3 27B IT· LMArena data1,156.4
114Mistral Medium 32505· LMArena data1,153.9
115GPT-5 nanoHigh· LMArena data1,152.2
116GLM 4.5V· LMArena data1,150.7
117Gemini 2.0 Flash001· LMArena data1,148
118Llama 4 Maverick· LMArena data1,142.8
119Claude 3.5 Sonnet20241022· LMArena data1,139.1
120Mistral Small 3.22506· LMArena data1,133.7
121Mistral Small 3.1 24B Instruct2503· LMArena data1,131.5
122Llama 4 Scout· LMArena data1,116.5
123Claude Haiku 3.520241022· LMArena data1,089.8
124Molmo 2 8B· LMArena data1,058.5

Showing 101–124 of 124 results.

Data credits: LMArena, CC BY 4.0 (licence)