Roboflow Vision Evals — OCR (September)
Multimodal
Mean similarity of OCR transcriptions to ground truth, preserving capitalization, symbols and layout. One or three runs per effort tier; atomic sample count unreported.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 94.0 |
| Claude Fable 5.1 | 94.0 |
| Opus 4.8 | 93.8 |
| Muse Spark 1.2 | 93.6 |
| Grok 4.7 | 93.5 |
| Claude Opus 5 | 93.2 |
| Kimi K3 | 93.0 |
| Muse Spark 1.1 | 92.8 |
| Gemini 3.1 Pro Preview | 92.6 |
| Grok 4.5 | 92.3 |
| Muse Glimmer 30B | 92.1 |
| GPT-6 Astra | 91.9 |
| GPT-6 Sol | 91.9 |
| Grok 4.6 | 91.8 |
| Sonnet 5 | 91.7 |
| GPT-5.5 | 91.7 |
| GPT-5.6 Luna | 91.5 |
| Qwen3.8 Flash | 91.3 |
| Muse Spark 1.3 | 91.3 |
| MiMo-V2.6-Pro | 90.7 |
| GPT-5.6 Sol | 90.7 |
| GLM-5.3 Flash | 90.6 |
| GPT-5.4 Mini | 89.5 |
| Gemini 3.6 Flash | 89.5 |
| GPT-5.6 Terra | 89.4 |
| Gemini 3.5 Flash | 89.3 |
| GLM 5V Turbo | 89.3 |
| Gemini 3.7 Flash | 89.0 |
| Gemini 3.8 Flash | 88.8 |
| Gemini 2.5 Pro | 88.8 |
| GPT-6 Luna | 88.5 |
| Kimi K2.6 | 88.5 |
| Qwen3-VL-235B-A22B-Instruct | 88.1 |
| Claude Opus 5.5 | 87.8 |
| Gemini 3 Flash Preview | 87.6 |
| Gemini 3.5 Flash-Lite | 87.4 |
| MiMo-V2.6-Flash | 87.0 |
| Qwen3.7-Plus | 86.5 |
| Qwen3.5-27B | 84.5 |
| Qwen3.7 Flash | 84.1 |