Embodied Household Navigation on ALFWorld
97Success RateGemini-3-Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-3-FlashProtocol=Prompting2026.05 | 97 | |
| ICRLBackbone=Qwen3-4B2026.05 | 92.5 | |
| Critique-GRPOBackbone=Qwen3-4B2026.05 | 91.5 | |
| ScalingInterBackbone=Qwen3-8B2026.05 | 91 | |
| Critique-GRPOBackbone=Qwen3-8B2026.05 | 90.5 | |
| ICRLBackbone=Qwen3-8B2026.05 | 90.5 | |
| ScalingInterBackbone=Qwen3-4B2026.05 | 90 | |
| MATPOBackbone=Qwen3-8B2026.05 | 89 | |
| MATPOBackbone=Qwen3-4B2026.05 | 87.5 | |
| GSPOBackbone=Qwen3-8B2026.05 | 83 | |
| GRPOBackbone=Qwen3-8B2026.05 | 82.5 | |
| GSPOBackbone=Qwen3-4B2026.05 | 82 | |
| GRPOBackbone=Qwen3-4B2026.05 | 81 | |
| Gemini-2.5-flashProtocol=Prompting2026.05 | 78.5 | |
| Qwen3-30B-A3BProtocol=Prompting2026.05 | 45.5 | |
| Qwen3-8BProtocol=Prompting2026.05 | 38 | |
| Qwen3-4BProtocol=Prompting2026.05 | 22 |