Agent Return Volatility on CAGE-2 v1 (test)
2SDFORGE w/o grad
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| FORGE w/o gradBackbone=Grok-4-Fast, Condition=Examples2026.05 | 2 | 96 | 96 | 96 | |
| FORGE w/o gradBackbone=Qwen3-235B, Condition=Examples2026.05 | 2.3 | 97 | 97 | 94 | |
| FORGE w/o gradBackbone=Qwen3-235B, Condition=Rules2026.05 | 2.6 | 97 | 97 | 88 | |
| FORGE w/o gradBackbone=Qwen3-235B, Condition=Mixed2026.05 | 2.8 | 97 | 97 | 86 | |
| FORGE w/o gradBackbone=Llama-4-Maverick, Condition=Mixed2026.05 | 9.3 | 89 | 77 | 64 | |
| FORGE w/o gradBackbone=Grok-4-Fast, Condition=Rules2026.05 | 15.1 | 73 | 76 | 42 | |
| FORGEBackbone=Llama-4-Maverick, Condition=Examples2026.05 | 15.7 | 81 | 74 | — | |
| FORGE w/o gradBackbone=Grok-4-Fast, Condition=Mixed2026.05 | 19.4 | 65 | 75 | 47 | |
| FORGEBackbone=Qwen3-235B, Condition=Mixed2026.05 | 20.4 | 77 | 77 | — | |
| FORGEBackbone=Qwen3-235B, Condition=Rules2026.05 | 21 | 76 | 75 | — | |
| FORGEBackbone=Gemini-2.5-Flash-Lite, Condition=Examples2026.05 | 21.1 | 61 | 65 | — | |
| FORGEBackbone=Llama-4-Maverick, Condition=Mixed2026.05 | 25.6 | 69 | 37 | — | |
| FORGEBackbone=Grok-4-Fast, Condition=Rules2026.05 | 26.1 | 53 | 58 | — | |
| FORGE w/o gradBackbone=Gemini-2.5-Flash-Lite, Condition=Rules2026.05 | 26.5 | 51 | 56 | 28 | |
| FORGE w/o gradBackbone=Gemini-2.5-Flash-Lite, Condition=Examples2026.05 | 27.2 | 50 | 55 | 29 | |
| FORGEBackbone=Gemini-2.5-Flash-Lite, Condition=Mixed2026.05 | 28.1 | 48 | 62 | — | |
| FORGE w/o gradBackbone=Gemini-2.5-Flash-Lite, Condition=Mixed2026.05 | 30.7 | 43 | 59 | 9 | |
| FORGE w/o gradBackbone=Llama-4-Maverick, Condition=Examples2026.05 | 35.3 | 57 | 41 | 125 | |
| FORGEBackbone=Qwen3-235B, Condition=Examples2026.05 | 35.8 | 59 | 48 | — | |
| FORGEBackbone=Grok-4-Fast, Condition=Mixed2026.05 | 36.9 | 33 | 52 | — | |
| FORGEBackbone=Gemini-2.5-Flash-Lite, Condition=Rules2026.05 | 37 | 31 | 39 | — | |
| ReflexionBackbone=Llama-4-Maverick, Condition=Mixed2026.05 | 40.6 | 50 | — | — | |
| FORGEBackbone=Grok-4-Fast, Condition=Examples2026.05 | 43.3 | 21 | 17 | — | |
| FORGEBackbone=Llama-4-Maverick, Condition=Rules2026.05 | 46.6 | 43 | 24 | — | |
| ReflexionBackbone=Grok-4-Fast, Condition=Examples2026.05 | 52.4 | 5 | — | — | |
| Zero-ShotBackbone=Gemini-2.5-Flash-Lite, Condition=No adaptation2026.05 | 53.9 | — | — | — | |
| FORGE w/o gradBackbone=Llama-4-Maverick, Condition=Rules2026.05 | 54.7 | 33 | 10 | 17 | |
| Zero-ShotBackbone=Grok-4-Fast, Condition=No adaptation2026.05 | 55.2 | — | — | — | |
| ReflexionBackbone=Llama-4-Maverick, Condition=Examples2026.05 | 59.7 | 27 | — | — | |
| ReflexionBackbone=Gemini-2.5-Flash-Lite, Condition=Examples2026.05 | 60.4 | 12 | — | — | |
| ReflexionBackbone=Gemini-2.5-Flash-Lite, Condition=Rules2026.05 | 60.5 | 12 | — | — | |
| ReflexionBackbone=Llama-4-Maverick, Condition=Rules2026.05 | 61 | 25 | — | — | |
| ReflexionBackbone=Grok-4-Fast, Condition=Rules2026.05 | 62.6 | 14 | — | — | |
| ReflexionBackbone=Qwen3-235B, Condition=Examples2026.05 | 69.5 | 20 | — | — | |
| ReflexionBackbone=Gemini-2.5-Flash-Lite, Condition=Mixed2026.05 | 74.6 | 38 | — | — | |
| ReflexionBackbone=Grok-4-Fast, Condition=Mixed2026.05 | 76.8 | 39 | — | — | |
| Zero-ShotBackbone=Llama-4-Maverick, Condition=No adaptation2026.05 | 81.5 | — | — | — | |
| ReflexionBackbone=Qwen3-235B, Condition=Rules2026.05 | 83.5 | 4 | — | — | |
| Zero-ShotBackbone=Qwen3-235B, Condition=No adaptation2026.05 | 87.3 | — | — | — | |
| ReflexionBackbone=Qwen3-235B, Condition=Mixed2026.05 | 89.4 | 2 | — | — |