Roboflow Vision Evals — Identification, Strict (September)
Multimodal
Normalized exact-match visual identification accuracy, reported beside judged accuracy under the September observed protocol.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3.1 Pro Preview | 100.0 |
| Gemini 3.6 Flash | 99.0 |
| Gemini 3.5 Flash | 99.0 |
| Claude Fable 5 | 96.9 |
| Gemini 3.8 Flash | 95.8 |
| Gemini 3.7 Flash | 95.8 |
| Claude Fable 5.1 | 93.8 |
| Muse Spark 1.3 | 92.7 |
| Gemini 3 Flash Preview | 90.6 |
| Qwen3-VL-235B-A22B-Instruct | 90.6 |
| Muse Spark 1.1 | 89.6 |
| Muse Spark 1.2 | 89.6 |
| Claude Opus 5.5 | 87.5 |
| Gemini 2.5 Pro | 87.5 |
| GPT-6 Sol | 87.5 |
| GPT-6 Astra | 86.5 |
| GPT-5.6 Sol | 86.5 |
| GPT-5.5 | 86.5 |
| Claude Opus 5 | 84.4 |
| Qwen3.7-Plus | 84.4 |
| Qwen3.7 Flash | 84.4 |
| GLM 5V Turbo | 84.4 |
| Qwen3.8 Flash | 83.3 |
| Grok 4.7 | 81.3 |
| Grok 4.6 | 81.3 |
| Gemini 3.5 Flash-Lite | 81.3 |
| MiMo-V2.6-Flash | 81.3 |
| Sonnet 5 | 81.3 |
| Kimi K3 | 81.3 |
| GPT-5.6 Terra | 80.2 |
| Grok 4.5 | 80.2 |
| GPT-5.4 Mini | 79.2 |
| GLM-5.3 Flash | 78.1 |
| Muse Glimmer 30B | 78.1 |
| Qwen3.5-27B | 78.1 |
| GPT-5.6 Luna | 77.1 |
| GPT-6 Luna | 76.0 |
| MiMo-V2.6-Pro | 76.0 |
| Opus 4.8 | 75.0 |
| Kimi K2.6 | 62.5 |