OCRBench v2 — Explicit v2 of OCRBench (expanded task coverage); distinct version.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Qwen3.7 Plus Preview | Alibaba | 67.1 | 1 | 2026-08-23 |
| 2 | NVIDIA Nemotron 3 Nano 30B A3B | NVIDIA | 65.8 | 1 | 2026-05-22 |
| 3 | Qwen3.6 35B A3B | Alibaba | 65.5 | 1 | 2026-07-02 |
| 4 | Qwen3.5 35B A3B | Alibaba | 65.3 | 1 | 2026-07-02 |
| 5 | Gemini 3 Pro | 63.4 | 1 | 2026-05-22 | |
| 6 | NVIDIA Nemotron Nano 12B v2 VL | NVIDIA | 61.2 | 1 | 2026-05-22 |
| 7 | Gemini 2.5 Pro | 59.3 | 1 | 2026-05-22 | |
| 8 | GLM-4.6V-Flash (9B) | Z.ai | 59 | 1 | 2026-05-22 |
| 9 | Qwen3.5 9B | Alibaba | 58.7 | 1 | 2026-05-22 |
| 10 | Qwen2.5 Omni 7B | Alibaba | 57.8 | 1 | 2026-08-23 |
| 11 | Llama 3.1 Nemotron Nano VL 8B V1 | NVIDIA | 56.4 | 1 | 2026-05-22 |
| 12 | GPT-5 | OpenAI | 55.5 | 3 | 2026-08-10 |
| 13 | Gemma 4 12B | 51.8 | 1 | 2026-07-02 | |
| 14 | Gemini 1.5 Pro | 51.6 | 1 | 2026-05-22 | |
| 15 | GPT-5.2 | OpenAI | 50.5 | 1 | 2026-05-22 |
| 16 | Claude Opus 4.6 | Anthropic | 48.4 | 1 | 2026-05-22 |
| 17 | LFM2.5-VL-3B | Liquid AI | 47.5 | 1 | 2026-08-14 |
| 18 | MiniCPM-o-4.5 | OpenBMB | 44.9 | 1 | 2026-05-22 |
| 19 | Gemma 4 E2B | 44.7 | 1 | 2026-08-12 | |
| 20 | Molmo2-8B | AllenAI | 44.7 | 1 | 2026-05-22 |
| 21 | LFM2-VL-3B (3.1B) | Liquid AI | 43.9 | 1 | 2026-08-12 |
| 22 | Ministral 3 14B | Mistral | 43.3 | 1 | 2026-05-22 |
| 23 | MiniCPM-V 4.6 | OpenBMB | 40.4 | 1 | 2026-07-09 |
| 24 | Phi 4 Multimodal Instruct | Microsoft | 38.1 | 2 | 2026-06-19 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.