metaai·lightalo unofficial · independent
Universe / Language & LLMs / Code World Model
Language & LLMs · 2025

Code World Model

Code World Model (CWM)

A neural debugger that simulates code in its head

open source 32B dense params

Latest: CWM 32B (Sept/Oct 2025), checkpoints at mid-train, SFT and RL stages

A 32B dense open-weights research LLM from Meta FAIR (released Sept–Oct 2025) that learns a 'world model' of code execution: mid-trained on observation-action trajectories from Python interpreters and agentic Docker environments, then multi-task RL on verifiable coding, math, and software-engineering tasks, with a 131k-token context.

Why it matters

CWM is the first open-weights LLM trained to model code execution itself — predicting variable states step by step like a neural debugger — and at 32B dense it reaches 65.8% on SWE-bench Verified. Releasing mid-train, SFT and RL checkpoints made the whole post-training pipeline studyable, rare transparency at that scale.

Facts

Try it yourself

Lineage

Descends fromCode Llama
Led toMuse Code

See the whole family tree →

Sources

More in Language & LLMs

Llama 4Llama APILlama 3 / 3.1 / 3.2 / 3.3Large Concept ModelsByte Latent TransformerLlama Stack

Read the Language & LLMs story on the sky →

✦ Open on the map Explore Language & LLMs Quiz me