OCRBench-V2 (zh): leaderboard
OCRBench v2 Chinese subset: Enhanced benchmark for evaluating Large Multimodal Models on visual text localization and reasoning with Chinese text content.
Metric: Score (imported). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Around October 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.7 Plus | 67.1 | #92 |
| 2 | Qwen 3.5 9B | 64.1 | #379 |
| 3 | Gemini 3 Pro (Preview) | 63.8 | #66 |
| 4 | Qwen 3.6 Plus | 63.6 | #120 |
| 5 | Gemini 2.5 Pro | 62.2 | #152 |
| 6 | Qwen3 Omni 30B A3B Instruct | 60 | #383 |
| 7 | Gemini 3.1 Pro (Preview) | 58.2 | #58 |
| 8 | GPT-5.4 (xHigh) | 57.7 | #79 (GPT-5.4) |
| 9 | Claude Opus 4.6 (Max) | 54.9 | #63 (Claude Opus 4.6) |
| 10 | GPT-5.2 | 52.6 | #110 |
| 11 | Ministral 3 14B | 44 | #650 |
| 12 | Phi-4 Multimodal Instruct | 37.3 | #906 |
| 13 | Molmo2-8B | 21.1 | #543 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ocrbench-v2-zh · How It Works · Data refreshed daily, snapshot 2026-10-20.