Roboflow Vision Evals — OCR Structured JSON (October)
Multimodal
JSON field matching score in the October OCR suite; missing and extra fields both count against the model. Single preliminary run per effort tier; question count unreported.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 94.1 |
| GPT-6 Astra | 91.8 |
| Gemini 3.8 Flash | 90.9 |
| Gemini 3.7 Flash | 90.6 |
| GPT-6.1 Sol | 90.4 |
| Claude Fable 5.1 | 89.5 |
| Gemini 3.6 Flash | 89.3 |
| Claude Fable 5 | 89.2 |
| Claude Opus 5 | 88.7 |
| Gemini 3.5 Flash | 88.1 |
| Gemini 3 Flash Preview | 87.5 |
| Muse Spark 1.2 | 87.0 |
| Grok 4.7 | 86.4 |
| Muse Spark 1.1 | 86.1 |
| Grok 4.6 | 85.7 |
| GPT-6 Sol | 85.3 |
| Claude Haiku 5.5 | 85.3 |
| Gemini 3.1 Pro Preview | 83.9 |
| GPT-5.6 Sol | 83.9 |
| Claude Sonnet 5.5 | 83.8 |
| Opus 4.8 | 83.5 |
| Kimi K3 | 82.9 |
| GPT-5.5 | 82.6 |
| Gemini 3.5 Flash-Lite | 82.1 |
| Qwen3.7-Plus (source unspecified) | 81.3 |
| GPT-6 Luna | 81.0 |
| Grok 4.5 | 80.9 |
| Muse Glimmer 30B | 80.2 |
| Qwen3.8 Flash | 79.3 |
| GPT-5.4 Mini | 79.0 |
| Kimi K2.6 | 78.5 |
| Gemini 2.5 Pro | 78.3 |
| Sonnet 5 | 78.3 |
| GPT-5.6 Terra | 77.3 |
| GPT-5.6 Luna | 77.0 |
| Mistral Large 4 Preview | 76.4 |
| Qwen3.5-27B | 76.0 |
| Qwen3.7 Flash | 75.7 |
| GLM-5.3 Flash | 72.0 |
| Qwen3-VL-235B-A22B-Instruct | 71.4 |