Agent-Editing World Model: Rethinking World Modeling for LLM Agents
Researchers propose Agent-Editing World Model to revise noisy agent reasoning instead of predicting tool responses.
The paper introduces the Agent-Editing World Model, which edits noisy reasoning and actions rather than predicting high-entropy tool responses. An Action Judge separates Critical, Exploratory, and Noisy decisions, and State Revision rewrites contaminated continuations from the same history. Trained on Search, Terminal, and Software Engineering, it reaches 70.5% macro-F1, beating the strongest frontier baseline by 10.6 points. EditAct then improves average scores by 3.2–6.7 points, and rejection-sampling fine-tuning adds 2.2–2.6 points over Self-RFT.
- AEWM models task progress instead of reconstructing tool responses.
- Action Judge labels decisions Critical, Exploratory, or Noisy.
- Action Judge scores 70.5% macro-F1, 10.6 points above the best baseline.
- EditAct gains 3.2–6.7 points across six benchmarks and three backbones.
- AEWM-RFT beats Self-RFT by 2.2–2.6 points without online guidance.
Full article200 words · extracted from arxiv.org · click to collapse
Recent advances in large language models (LLMs) have enabled agents to tackle long-horizon tasks across diverse environments. To further improve agent performance, existing language world models typically predict environment observations, yet reconstructing high-entropy, execution-dependent tool responses offers limited value when real feedback is available. Meanwhile, agents suffer from \emph{task-state contamination}, where unsupported assumptions and outdated plans persist in history and distort subsequent decisions. We propose the \textbf{Agent-Editing World Model (AEWM)}, which models how reasoning and actions shape future task progress rather than simulating tool responses. AEWM combines \textbf{Action Judge} to distinguish \textsc{Critical}, \textsc{Exploratory}, and \textsc{Noisy} decisions with \textbf{State Revision} to edit noisy reasoning--action continuations from the same observed history. \textbf{EditAct} integrates these capabilities with real execution, directly changing the state underlying subsequent decisions rather than merely providing critiques. We train AEWM across Search, Terminal, and Software Engineering through mid-training and supervised fine-tuning. AEWM achieves 70.5\% macro-F1 on our Action Judge benchmark, exceeding the strongest frontier baseline by 10.6 points. Across six benchmarks and three agent backbones, EditAct improves average scores by 3.2--6.7 points over the strongest baseline. Furthermore, rejection sampling fine-tuning on verified EditAct trajectories, termed \textbf{AEWM-RFT}, improves over Self-RFT by 2.2--2.6 points across three domains without online AEWM guidance.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.28416