Visual Perception on VStar (test)
92.7AccuracyZwZ-4B
Evaluation Results
| Method | Links | |
|---|---|---|
| ZwZ-4BInference Strategy=Single Forward Pass, Model Scale=4B2026.02 | 92.7 | |
| SenseNova-MARS-8BInference Strategy=Agentic, Model Scale=8B2026.02 | 92.2 | |
| ZwZ-8BInference Strategy=Single Forward Pass, Model Scale=8B2026.02 | 91.1 | |
| DeepEyesInference Strategy=Agentic, Model Scale=7B2026.02 | 90.1 | |
| Qwen3-VL-8B+toolInference Strategy=Tool Use, Model Scale=8B2026.02 | 90.1 | |
| ZwZ-7BInference Strategy=Single Forward Pass, Model Scale=7B2026.02 | 88.5 | |
| Mini o3Inference Strategy=Agentic, Model Scale=7B2026.02 | 88.2 | |
| Skywork-R1V4Inference Strategy=Agentic, Model Scale=30B2026.02 | 88 | |
| Qwen3-VL-4B+toolInference Strategy=Tool Use, Model Scale=4B2026.02 | 88 | |
| Qwen3-VL-8BInference Strategy=Single Forward Pass, Model Scale=8B2026.02 | 86.4 | |
| Pixel-ReasonerInference Strategy=Agentic, Model Scale=7B2026.02 | 84.3 | |
| ThymeInference Strategy=Agentic, Model Scale=7B2026.02 | 82.2 | |
| DeepEyesV2Inference Strategy=Agentic, Model Scale=7B2026.02 | 81.8 | |
| Qwen3-VL-4BInference Strategy=Single Forward Pass, Model Scale=4B2026.02 | 80.1 | |
| Qwen2.5-VL-7BInference Strategy=Single Forward Pass, Model Scale=7B2026.02 | 78.5 |