PerceptionBench - Comparison
Multimodal · 2026-07-16
The PerceptionBench comparison category contains 279 questions that require comparing visible quantities or properties across image elements.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 67.0 |
| GPT-5.5 | 65.6 |
| Gemini 3.1 Pro Preview | 61.7 |
| Doubao Seed 2.1 Pro | 59.5 |
| Kimi K3 | 59.1 |
| Claude Fable 5 | 56.1 |
| Qwen3.7-Plus | 55.9 |
| Gemini 3.5 Flash | 54.8 |
| Qwen3.5 397B A17B | 49.8 |
| Opus 4.8 | 48.4 |
| GLM 5V Turbo | 45.2 |
| Gemma 4 31B IT | 43.7 |
| Grok 4.5 | 43.7 |
| Kimi K2.6 | 39.1 |
| GLM-4.6V | 34.8 |
| MiniMax M3 | 31.2 |