Benchmark
OCR Arena Elo
OCR benchmarks measure how well models read text in images and documents. Higher scores are better.
- Models measured
- 11
- Scores recorded
- 17
- Sources
- 1
- IQ weight
- 0.07
Score distribution
- 1,543.3
- Mean
- 1,525
- Median
- 168.3
- Std dev
- 1,365
- 10th percentile
- 1,751
- 90th percentile
- 1,331
- Min
This benchmark has a weight of 0.07 for chat models, measured over 9 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash· OCR Arena data | 1,793 |
| 2 | Claude Opus 4.5low· OCR Arena data | 1,778 |
| 3 | Gemini 2.5 Pro· OCR Arena data | 1,751 |
| 4 | Gemini 2.5 Flash· OCR Arena data | 1,730 |
| 5 | Claude Opus 4.6adaptive· OCR Arena data | 1,679 |
| 6 | Claude Opus 4.6medium· OCR Arena data | 1,666 |
| 7 | Claude Opus 4.5medium· OCR Arena data | 1,654 |
| 8 | Claude Opus 4.6low· OCR Arena data | 1,644 |
| 9 | Qwen: Qwen3 VL 8B Thinking· OCR Arena data | 1,525 |
| 10 | GLM OCR· OCR Arena data | 1,448 |
| 11 | GPT 5low· OCR Arena data | 1,406 |
| 12 | DeepSeek-OCR· OCR Arena data | 1,377 |
| 13 | GPT 5.1medium· OCR Arena data | 1,367 |
| 14 | GPT 5medium· OCR Arena data | 1,366 |
| 15 | GPT 5.1low· OCR Arena data | 1,365 |
| 16 | GPT 5.2none· OCR Arena data | 1,356 |
| 17 | GPT 5.2medium· OCR Arena data | 1,331 |