CWM: An Open-Weights LLM for Research on Code Generation with World Models

2025-09-30

AI Summary

This work addresses the limited reasoning and planning capabilities in code generation by proposing Code World Model (CWM), the first open-source 32B-parameter large language model to integrate the world model paradigm into code intelligence. CWM undergoes mid-training via agent-based Docker environments, jointly executing Python interpreter observations and actions to learn executable trajectories; it further combines supervised fine-tuning with multi-task reinforcement learning—covering verifiable coding, mathematical reasoning, and multi-turn software engineering tasks. Its pure-decoder architecture supports a 131K-context window, enabling stepwise simulation of dynamic execution environments and causal reasoning. Experiments demonstrate substantial improvements over baselines: 65.8% pass@1 on SWE-bench Verified, 68.6% on LiveCodeBench, 96.6% on Math-500, and 76.0% on AIME 2024. These results validate the efficacy of world modeling for code-level reasoning. All training checkpoints are publicly released to advance research in agentic programming.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

We release Code World Model (CWM), a 32-billion-parameter open-weights LLM, to advance research on code generation with world models. To improve code understanding beyond what can be learned from training on static code alone, we mid-train CWM on a large amount of observation-action trajectories from Python interpreter and agentic Docker environments, and perform extensive multi-task reasoning RL in verifiable coding, math, and multi-turn software engineering environments. With CWM, we provide a strong testbed for researchers to explore the opportunities world modeling affords for improving code generation with reasoning and planning in computational environments. We present first steps of how world models can benefit agentic coding, enable step-by-step simulation of Python code execution, and show early results of how reasoning can benefit from the latter. CWM is a dense, decoder-only LLM trained with a context size of up to 131k tokens. Independent of its world modeling capabilities, CWM offers strong performance on general coding and math tasks: it reaches pass@1 scores of 65.8% on SWE-bench Verified (with test-time scaling), 68.6% on LiveCodeBench, 96.6% on Math-500, and 76.0% on AIME 2024. To support further research on code world modeling, we release model checkpoints after mid-training, SFT, and RL.

Problem

• Advancing code generation research using world models and large language models
• Improving code understanding through interpreter and agentic environment training
• Enabling step-by-step code execution simulation and reasoning in computational environments

Innovation

• Mid-training on Python interpreter trajectories for code understanding
• Multi-task reasoning RL in verifiable coding environments
• World modeling enables step-by-step Python code execution simulation