Agentic Task Evaluation on VitaBench In-store
34.62Avg@2Snapshot
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SnapshotBackbone=Qwen3-30B-A3B2026.05 | 34.62 | 50 | |
| PPO-EWMABackbone=Qwen3-30B-A3B2026.05 | 33.41 | 48 | |
| Linear_proxBackbone=Qwen3-30B-A3B2026.05 | 31.47 | 47 | |
| SnapshotBackbone=Qwen3-4B2026.05 | 28.89 | 47 | |
| PPO-EWMABackbone=Qwen3-4B2026.05 | 25 | 50 | |
| Linear_proxBackbone=Qwen3-4B2026.05 | 22.37 | 40 | |
| Decoupled PPOBackbone=Qwen3-4B2026.05 | 19.83 | 37 | |
| Decoupled PPOBackbone=Qwen3-30B-A3B2026.05 | 18.28 | 32 |