PerceptionBench - OCR
Multimodal · 2026-07-16
The PerceptionBench OCR category contains 255 questions that test exact reading of visible text, including case, spacing, punctuation, and localized text regions.
Top models (higher is better)
| Model | Score |
|---|---|
| Doubao Seed 2.1 Pro | 66.7 |
| Claude Fable 5 | 64.3 |
| Gemini 3.1 Pro Preview | 64.3 |
| Kimi K3 | 61.2 |
| Gemini 3.5 Flash | 59.6 |
| GPT-5.5 | 56.5 |
| GPT-5.6 Sol | 54.9 |
| Qwen3.7-Plus | 54.5 |
| Opus 4.8 | 54.1 |
| Qwen3.5 397B A17B | 52.9 |
| Gemma 4 31B IT | 46.7 |
| Grok 4.5 | 43.9 |
| GLM 5V Turbo | 43.5 |
| Kimi K2.6 | 40.8 |
| GLM-4.6V | 39.2 |
| MiniMax M3 | 35.7 |