mlsanityData to 7 Oct 2026

World-Model Collapse as a Phase Transition

arXiv 2606.31399 ↗First posted 30 Jun 2026cs.AIcited by 1

On the map: World models → World models for language agents

also in 1 more
Abstract (arXiv)

Water looks unchanged as it warms, then at a critical point it boils. We ask whether long-horizon language agents show an analogous transition in their implicit world models. In some parameter settings, changing state load by a small amount, or adding a single step of horizon, leaves behavior nearly unchanged; near a critical boundary, the same small change causes a sudden world collapse. We study this effect in a deterministic task family with exact per-step gold state. A large grid search over state cardinality, dependency density, horizon, branching, observation mode, and mutation rate reveals a phase diagram: a solved plateau, a narrow transition band, and a collapse floor. Per-step traces show the mechanism: world-state fidelity fails before action validity, so the agent is not merely choosing a bad action; it is acting from a corrupted world. Stronger models translate the critical boundary but do not remove the qualitative transition. These results make world-model collapse a measurable bottleneck for long-horizon agents.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in World models for language agents, one of 11 directions in World models. Builds on 65 of its 76 references that are in the map. First posted on arXiv 30 Jun 2026.

Builds on

65 of 76 references.

Closest ancestors · 2

Same direction, newest first

  1. 2023Generative Agents: Interactive Simulacra of Human Behavior7 Apr
  2. 2020ALFWorld: Aligning Text and Embodied Environments for Interactive Learning8 Oct

Other references · 63

  1. 2024The Llama 3 Herd of Modelscited by 14,192
  2. 2020Scaling Laws for Neural Language Modelscited by 6,411
  3. 2024GPT-4o System Cardcited by 4,467
  4. 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,072
  5. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,064
  6. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,383
  7. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,193
  8. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Modelscited by 3,155
  9. 2022Emergent Abilities of Large Language Modelscited by 2,483
  10. 2023A Survey on Large Language Model based Autonomous Agentscited by 2,203
  11. 2023WebArena: A Realistic Web Environment for Building Autonomous Agentscited by 1,908
  12. 2023ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIscited by 1,880
38 more
  1. 2024SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineeringcited by 1,628
  2. 2018World Modelscited by 1,444
  3. 2023MemGPT: Towards LLMs as Operating Systemscited by 1,295
  4. 2024τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domainscited by 1,280
  5. 2023AgentBench: Evaluating LLMs as Agentscited by 1,211
  6. 2022WebShop: Towards Scalable Real-World Web Interaction with Grounded Language Agentscited by 1,197
  7. 2023Mastering Diverse Domains through World Modelscited by 1,193
  8. 2023GAIA: a benchmark for General AI Assistantscited by 1,084
  9. 2020Mastering Atari with Discrete World Modelscited by 990
  10. 2022In-context Learning and Induction Headscited by 930
  11. 2018Reconciling modern machine learning practice and the bias-variance trade-offcited by 919
  12. 2023Graph of Thoughts: Solving Elaborate Problems with Large Language Modelscited by 873
  13. 2023Progress measures for grokking via mechanistic interpretabilitycited by 822
  14. 2024Genie: Generative Interactive Environmentscited by 706
  15. 2019Deep Double Descent: Where Bigger Models and More Data Hurtcited by 667
  16. 2022Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasetscited by 599
  17. 2023Are Emergent Abilities of Large Language Models a Mirage?cited by 515
  18. 2023LLM+P: Empowering Large Language Models with Optimal Planning Proficiencycited by 496
  19. 2018A Mean Field View of the Landscape of Two-Layers Neural Networkscited by 463
  20. 2023Cognitive Architectures for Language Agentscited by 422
  21. 2024TravelPlanner: A Benchmark for Real-World Planning with Language Agentscited by 385
  22. 2022PlanBench: An Extensible Benchmark for Evaluating Large Language Models on Planning and Reasoning about Changecited by 372
  23. 2022ScienceWorld: Is your Agent Smarter than a 5th Grader?cited by 298
  24. 2025MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agentscited by 226
  25. 2024LLMs Can't Plan, But Can Help Planning in LLM-Modulo Frameworkscited by 178
  26. 2021The Principles of Deep Learning Theorycited by 169
  27. 2023Generalized Planning in PDDL Domains with Pretrained Large Language Modelscited by 139
  28. 2024On the Self-Verification Limitations of Large Language Models on Reasoning and Planning Taskscited by 137
  29. 2024StableToolBench: Towards Stable Large-Scale Benchmarking on Tool Learning of Large Language Modelscited by 129
  30. 2023On the Planning Abilities of Large Language Models (A Critical Investigation with a Proposed Benchmark)cited by 85
  31. 2024Can Large Language Models Reason and Plan?cited by 82
  32. 2025Look Back to Reason Forward: Revisitable Memory for Long-Context LLM Agentscited by 37
  33. 2026STALE: Can LLM Agents Know When Their Memories Are No Longer Valid?cited by 36
  34. 2026WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluationcited by 29
  35. 2023Dissociating language and thought in large language modelscited by 27
  36. 2025UltraHorizon: Benchmarking Agent Capabilities in Ultra Long-Horizon Scenarioscited by 22
  37. 2024Revealing the Barriers of Language Agents in Planningcited by 20
  38. 2025WorldPrediction: A Benchmark for High-level World Modeling and Long-horizon Procedural Planningcited by 19

50 most cited of 63 All 63 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Dude, Where's My State? Execution Information Requirements for Stateful Agentscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026World Model Science: Self-Organized Criticality, Weak Chaos, and Metastable Belief Dynamics in Long-Horizon LLM Agents
  2. 2025Why Do LLM Agents Fail in Exploring New Environments? A World-Modeling Perspective
  3. 2026Long-Horizon Textual World Modeling through Structured Reasoning
  4. 2026Why We Need World Models for AGI: Where LLMs Fail and How World Models May Outperform
  5. 2026The Long-Horizon Task Mirage? Diagnosing Where and Why Agentic Systems Break
  6. 2025Limits of Emergent Reasoning of Large Language Models in Agentic Frameworks for Deterministic Games
  7. 2026World Agent: Can Language Models Keep a World Running?
  8. 2026DWM: Separating World Effects from Actions in Latent World Models