Grid-world Navigation on FrozenLake Implicit Hidden Drift
88.8Success Rate (Source)Qwen3-30B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-30BBackbone=Qwen3-30B, GLOVE Augmentation=false2026.01 | 88.8 | 46.9 | — | |
| Qwen3-30B + GLOVEBackbone=Qwen3-30B, GLOVE Augmentation=true2026.01 | 85 | 47.5 | 0.6 | |
| GPT-4oBackbone=GPT-4o, GLOVE Augmentation=false2026.01 | 60 | 51.9 | — | |
| GPT-4o + GLOVEBackbone=GPT-4o, GLOVE Augmentation=true2026.01 | 51.2 | 96.2 | 44.3 | |
| Qwen2.5-7B + GLOVEBackbone=Qwen2.5-7B, GLOVE Augmentation=true2026.01 | 41.9 | 100 | 2.5 | |
| Qwen2.5-7BBackbone=Qwen2.5-7B, GLOVE Augmentation=false2026.01 | 40.6 | 97.5 | — | |
| Llama3.3-70B + GLOVEBackbone=Llama3.3-70B, GLOVE Augmentation=true2026.01 | 40 | 97.5 | 1.3 | |
| Llama3.3-70BBackbone=Llama3.3-70B, GLOVE Augmentation=false2026.01 | 39.4 | 96.2 | — | |
| Grok-3Backbone=Grok-3, GLOVE Augmentation=false2026.01 | 35.6 | 68.1 | — | |
| Grok-3 + GLOVEBackbone=Grok-3, GLOVE Augmentation=true2026.01 | 33.8 | 98.8 | 30.7 | |
| DeepSeek-V3.2 + GLOVEBackbone=DeepSeek-V3.2, GLOVE Augmentation=true2026.01 | 27.5 | 95.6 | 20 | |
| DeepSeek-V3.2Backbone=DeepSeek-V3.2, GLOVE Augmentation=false2026.01 | 26.9 | 75.6 | — | |
| Llama3.1-8BBackbone=Llama3.1-8B, GLOVE Augmentation=false2026.01 | 13.1 | 52.5 | — | |
| Llama3.1-8B + GLOVEBackbone=Llama3.1-8B, GLOVE Augmentation=true2026.01 | 12.5 | 87.5 | 35 |