Roboflow Vision Evals — OCR Text Localization (October)
Multimodal
Located-text matching score in the October OCR suite: text must match exactly and its box must overlap ground truth with IoU above 0.5. Single preliminary run per effort tier; question count unreported.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 76.8 |
| GPT-6.1 Sol | 76.7 |
| GPT-6 Sol | 68.7 |
| Gemini 3.6 Flash | 62.1 |
| Claude Opus 5.5 | 61.1 |
| Claude Sonnet 5.5 | 60.6 |
| Gemini 3.7 Flash | 59.2 |
| Gemini 3.8 Flash | 58.6 |
| Gemini 3.5 Flash | 56.2 |
| GPT-5.6 Sol | 52.2 |
| Gemini 3.1 Pro Preview | 52.0 |
| GPT-6 Luna | 50.0 |
| Claude Fable 5.1 | 46.5 |
| Gemini 3.5 Flash-Lite | 46.4 |
| Claude Haiku 5.5 | 45.8 |
| GPT-5.6 Luna | 43.8 |
| Kimi K3 | 43.4 |
| Claude Opus 5 | 42.6 |
| GPT-5.6 Terra | 42.3 |
| Qwen3.5-27B | 40.3 |
| GLM 5V Turbo | 38.7 |
| Qwen3.8 Flash | 37.4 |
| Gemini 3 Flash Preview | 36.9 |
| Gemini 2.5 Pro | 35.9 |
| Muse Spark 1.1 | 33.9 |
| Claude Fable 5 | 33.3 |
| Qwen3.7 Flash | 31.7 |
| Muse Spark 1.2 | 31.5 |
| GPT-5.5 | 30.5 |
| Qwen3.7-Plus (source unspecified) | 30.4 |
| Kimi K2.6 | 29.8 |
| Mistral Large 4 Preview | 28.8 |
| Opus 4.8 | 24.2 |
| Qwen3-VL-235B-A22B-Instruct | 24.0 |
| MiMo-V2.6-Pro | 18.7 |
| GLM-5.3 Flash | 18.6 |
| Grok 4.7 | 14.8 |
| Muse Glimmer 30B | 14.8 |
| MiMo-V2.6-Flash | 13.2 |
| Sonnet 5 | 13.1 |