Vision-Language Modeling on 16 VLM benchmark suite macro-average
75.5PASS@1 AccuracyQwen3.5-27B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3.5-27BRole=Teacher2026.06 | 75.5 | — | |
| Qwen3.5-9BTraining Stage=ZPPO2026.06 | 73 | 2.8 | |
| Qwen3.5-4BTraining Stage=ZPPO2026.06 | 70.9 | 4 | |
| Qwen3.5-9BTraining Stage=Base2026.06 | 70.2 | — | |
| Qwen3.5-4BTraining Stage=Base2026.06 | 66.9 | — | |
| Qwen3.5-2BTraining Stage=ZPPO2026.06 | 62 | 5.2 | |
| Qwen3.5-2BTraining Stage=Base2026.06 | 56.8 | — | |
| Qwen3.5-0.8BTraining Stage=ZPPO2026.06 | 50.3 | 9.3 | |
| Qwen3.5-0.8BTraining Stage=Base2026.06 | 41 | — |