Meta AI Releases Open-Weights Model for Code Generation Research
The 32-billion-parameter Code World Model was trained on code execution data to improve its understanding of how software behaves when it runs.
An Open Approach to Code Generation
Meta AI Research released a new open-weights large language model on September 24, 2025, specifically designed to advance research on code generation. The model, named the Code World Model (CWM), has 32 billion parameters and is being made available to academic and commercial researchers. This move continues Meta's strategy of fostering an open-source ecosystem around its AI tools, a distinct path from the closed, proprietary approach of competitors like Alibaba and OpenAI.
CWM is a dense, decoder-only model trained with a large context size of up to 131,000 tokens. A key innovation in …
Archive Access
This article is older than 24 hours. Create a free account to access our 7-day archive.