CC-OCR
Multimodal · 2024-12-03
CC-OCR is a comprehensive OCR benchmark for large multimodal models covering multi-scene text reading, multilingual text reading, document parsing and key information extraction. Release tables report its overall score, an average over the four tracks whose metrics include edit-distance and F1 measures.
Top models (higher is better)
| Model | Score |
|---|---|
| Qwen3.5-Omni-Plus | 83.4 |
| Qwen3.6 Plus (2026-04-02) | 83.4 |
| Qwen3.5 Plus (2026-02-15) | 83.0 |
| Qwen3.5 397B A17B | 82.0 |
| Qwen3.6 35B A3B | 81.9 |
| Qwen3.5 122B A10B | 81.8 |
| Qwen3.6 27B | 81.2 |
| Qwen3.5-27B | 81.0 |
| Qwen3.5-Omni-Flash | 80.8 |
| Qwen3.5 35B A3B | 80.7 |
| Kimi K2.5 | 79.7 |
| Qwen3.5-9B | 79.3 |
| Gemini 3 Pro Preview | 79.0 |
| Opus 4.5 | 76.9 |
| Qwen3.5-4B | 76.7 |
| Qwen3.5-2B | 75.8 |
| Gemma 4 31B IT | 75.7 |
| Gemma 4 26B A4B IT | 74.5 |
| Gemini 2.5 Flash-Lite | 72.9 |
| GPT-5 Mini | 70.8 |
| GPT-5.2 | 70.3 |
| Sonnet 4.5 | 68.1 |
| Qwen3.5-0.8B | 66.7 |
| GPT-5 Nano | 58.9 |