PerceptionBench - Context Integration
Multimodal · 2026-07-16
The PerceptionBench context-integration category contains 255 questions that require combining visual evidence across regions or multiple images while remaining grounded in the presented content.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3.1 Pro Preview | 61.2 |
| Doubao Seed 2.1 Pro | 60.4 |
| GPT-5.6 Sol | 60.0 |
| Claude Fable 5 | 59.8 |
| GPT-5.5 | 58.0 |
| Gemini 3.5 Flash | 53.3 |
| Kimi K3 | 53.3 |
| Qwen3.7-Plus | 52.2 |
| Qwen3.5 397B A17B | 50.2 |
| Gemma 4 31B IT | 45.9 |
| Opus 4.8 | 44.7 |
| Kimi K2.6 | 40.4 |
| Grok 4.5 | 39.6 |
| GLM-4.6V | 33.7 |
| GLM 5V Turbo | 32.9 |
| MiniMax M3 | 31.0 |