List your company

Planning on AlfWorld, BabyAI, Maze, SciWorld, TextCraft, WebShop, Wordle (held-out)

17.5AlfWorld Success RateReAct
2.946.7210.514.28May 29, 2026

Evaluation results

MethodLinks
ReActplanner=shared one-step lookahead planner2026.0517.586.783.383.593.85610074.4
LLM-Directplanner=shared one-step lookahead planner, icl_prompt_context=k=3 transitions2026.0514.588.91008595.84610075.8
Word2Worldplanner=shared one-step lookahead planner2026.059.675.683.366.786.522.610063.5
PatchWorld-Simpleplanner=shared one-step lookahead planner, backbone=Qwen3-Coder-480B2026.05687.810086.595.858.510076.4
PatchWorld-Residualplanner=shared one-step lookahead planner, backbone=Qwen3-Coder-480B2026.055.586.183.382.596.95610072.9
WorldCoderplanner=shared one-step lookahead planner2026.053.581.766.788.567.74310064.4
PoE-Worldplanner=shared one-step lookahead planner2026.053.584.483.393.570.849.510069.3