PerceptionBench - Fine-Grained Recognition
Multimodal · 2026-07-16
The PerceptionBench fine-grained-recognition category contains 290 questions that test discrimination of subtle visual details and closely similar image elements.
Top models (higher is better)
| Model | Score |
|---|---|
| Doubao Seed 2.1 Pro | 56.6 |
| GPT-5.6 Sol | 55.9 |
| Kimi K3 | 55.9 |
| Gemini 3.1 Pro Preview | 54.8 |
| Gemini 3.5 Flash | 51.7 |
| Claude Fable 5 | 51.6 |
| GPT-5.5 | 47.2 |
| Qwen3.7-Plus | 46.8 |
| Qwen3.5 397B A17B | 44.8 |
| Opus 4.8 | 40.7 |
| Gemma 4 31B IT | 39.0 |
| GLM 5V Turbo | 36.6 |
| Grok 4.5 | 36.2 |
| Kimi K2.6 | 34.5 |
| GLM-4.6V | 29.3 |
| MiniMax M3 | 26.6 |