Roboflow Vision Evals — Extraction, Judged (September)
Multimodal
Image data-extraction correctness judged against ground truth by Gemini 3.5 Flash at temperature zero. One or three runs per effort tier; atomic sample count unreported.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3.8 Flash | 97.3 |
| Gemini 3 Flash Preview | 96.9 |
| Gemini 3.7 Flash | 96.2 |
| Gemini 3.1 Pro Preview | 95.9 |
| Gemini 3.6 Flash | 95.9 |
| Gemini 3.5 Flash | 95.5 |
| Claude Fable 5.1 | 93.5 |
| Claude Opus 5.5 | 93.5 |
| Gemini 3.5 Flash-Lite | 91.8 |
| Claude Fable 5 | 91.8 |
| GPT-6 Astra | 91.1 |
| Sonnet 5 | 89.7 |
| Claude Opus 5 | 89.7 |
| Muse Spark 1.2 | 89.0 |
| Muse Spark 1.1 | 89.0 |
| Opus 4.8 | 88.7 |
| Muse Spark 1.3 | 88.7 |
| Qwen3-VL-235B-A22B-Instruct | 87.6 |
| Grok 4.7 | 87.6 |
| GPT-5.6 Sol | 86.9 |
| GPT-5.5 | 86.9 |
| Muse Glimmer 30B | 86.6 |
| Grok 4.6 | 85.6 |
| Qwen3.8 Flash | 84.9 |
| Gemini 2.5 Pro | 84.5 |
| Kimi K3 | 84.5 |
| GPT-5.4 Mini | 84.2 |
| Qwen3.7-Plus | 83.5 |
| GLM-5.3 Flash | 83.5 |
| Grok 4.5 | 83.5 |
| GPT-6 Sol | 82.5 |
| MiMo-V2.6-Flash | 82.5 |
| GPT-5.6 Luna | 81.8 |
| GLM 5V Turbo | 81.4 |
| MiMo-V2.6-Pro | 81.1 |
| GPT-5.6 Terra | 80.4 |
| Qwen3.5-27B | 79.4 |
| Kimi K2.6 | 79.4 |
| Qwen3.7 Flash | 77.3 |
| GPT-6 Luna | 68.0 |