OCRBench v2: leaderboard
OCRBench v2 evaluates large multimodal models on bilingual visual text localization and reasoning tasks.
Metric: Average (imported). Source: 99franklin.github.io. Source: benchmarklist.com. Saturation forecast: Around June 2027. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 9B | 64.1 | #379 |
| 2 | Gemini 3 Pro (Preview) | 63.8 | #66 |
| 3 | Gemini 2.5 Pro | 62.2 | #152 |
| 4 | Qwen3 Omni 30B A3B Instruct | 61.3 | #383 |
| 5 | Claude Opus 4.6 | 59.8 | #63 |
| 6 | GPT-5.2 | 52.6 | #110 |
| 7 | Gemini 1.5 Pro | 51.6 | #445 |
| 8 | Molmo2-8B | 44.7 | #543 |
| 9 | Ministral 3 14B | 44 | #650 |
| 10 | Phi-4 Multimodal Instruct | 38.1 | #906 |
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 · How It Works · Data refreshed daily, snapshot 2026-10-20.