Agentic task on VitaBench Delivery
30.74Avg@2Snapshot
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SnapshotBackbone=Qwen3-30B-A3B2026.05 | 30.74 | 45 | |
| PPO-EWMABackbone=Qwen3-30B-A3B2026.05 | 28.49 | 43 | |
| SnapshotBackbone=Qwen3-4B2026.05 | 27.33 | 42 | |
| PPO-EWMABackbone=Qwen3-4B2026.05 | 25.88 | 39 | |
| Decoupled PPOBackbone=Qwen3-30B-A3B2026.05 | 25.88 | 39 | |
| Linear_proxBackbone=Qwen3-30B-A3B2026.05 | 20.74 | 33 | |
| Decoupled PPOBackbone=Qwen3-4B2026.05 | 19.56 | 33 | |
| Linear_proxBackbone=Qwen3-4B2026.05 | 19.1 | 28 |