ALFWorld

Embodied text env

RL environmentAgentsTask completion rate134 items

What this measures

Embodied instruction-following: an agent reads commands like 'put a clean mug in the coffee machine' and acts in a simulated home, one step at a time. Tests grounded, sequential planning.

Sample

'Heat a slice of bread and put it on the counter.' — the agent issues navigate/pick/heat/place actions in a text world.

Leaderboard · 0 models

Task completion rate — higher is better

No scores recorded yet.

What improves this score

Datasets & environments shown to raise it.

No data yet.

Research using this benchmark

all →

LatentSkill: From In-Context Textual Skills to In-Weight Latent Skills for LLM Agents OPID: On-Policy Skill Distillation for Agentic Reinforcement Learning HarnessX: A Composable, Adaptive, and Evolvable Agent Harness Foundry Skill0.5: Joint Skill Internalization and Utilization for Out-of-Distribution Generalization in Agentic Reinforcement Learning TurnOPD: Making On-Policy Distillation Turn-Aware for Efficient Long-Horizon Agent Training

Matched by name in title/abstract.