Roboflow Vision Evals - OCR
Multimodal · 2026-07-22
The Roboflow Vision Evals OCR task measures mean transcription similarity to ground truth across 37 image samples.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 94.0 |
| Opus 4.8 | 93.8 |
| Kimi K3 | 93.0 |
| Gemini 3.1 Pro Preview | 92.6 |
| Muse Spark 1.1 | 92.5 |
| Sonnet 5 | 91.7 |
| GPT-5.5 | 91.2 |
| Gemini 3.5 Flash | 91.1 |
| GPT-5.6 Sol | 90.7 |
| GLM 5V Turbo | 89.3 |
| Gemini 2.5 Pro | 88.8 |
| GPT-5.6 Terra | 88.8 |
| Kimi K2.6 | 88.5 |
| Gemini 3.6 Flash | 88.4 |
| GPT-5.6 Luna | 88.4 |
| Qwen3-VL-235B-A22B-Instruct | 88.1 |
| GPT-5.4 Mini | 88.1 |
| Gemini 3 Flash Preview | 87.6 |
| Gemini 3.5 Flash-Lite | 87.4 |
| Qwen3.7-Plus | 86.5 |
| Qwen3.5-27B | 84.5 |