OCR benchmark
Higher is better
| # | Model | Score | Source |
|---|---|---|---|
| ★ | Claude 3.5 Sonnet (Oct 2024) | 91 | codesota-api |
| 2 | Qwen2.5-Coder-32B-Instruct | 90.2 | codesota-api |
| 3 | DeepSeek-Coder-V2-Instruct | 89.4 | codesota-api |
| 4 | claude-35-sonnet | 89.2 | codesota-api |
| 5 | gpt-4o | 87.8 | codesota-api |
| 6 | GPT-4o (Aug 2024) | 86.8 | codesota-api |
| 7 | Qwen2.5-Coder-7B-Instruct | 83.5 | codesota-api |
| 8 | Codestral 22B v0.1 | 78.2 | codesota-api |
| 9 | Llama 4 Maverick (17B-128E) | 77.6 | codesota-api |
| 10 | DeepSeek-V3 | 75.4 | codesota-api |
| 11 | Gemma 3 27B IT | 74.4 | codesota-api |
| 12 | Gemma 3 12B IT | 73 | codesota-api |
| 13 | Llama 4 Scout (17B-16E) | 67.8 | codesota-api |
| 14 | Gemma 3 4B IT | 63.2 | codesota-api |