# Code World Model

> Meta (Sep 2025) — LLM trained to reason about how code affects program state

- Category: [Foundation Models](https://ailandscape.org/category/foundation-models) › Coding Models
- Homepage: https://huggingface.co/facebook/cwm
- Tags: open-weights
- Added to the landscape: 2026-03-18

## Similar tools in Coding Models

- [Alibaba Cloud (Qwen)](https://ailandscape.org/tool/alibaba-cloud-qwen): Alibaba's Qwen family — Qwen3-Coder and Qwen3-Max lead open-weight coding models
- [ByteDance](https://ailandscape.org/tool/bytedance): ByteDance's Volcengine AI — Doubao-Seed-Code model integrated into the Trae IDE
- [Cognition (SWE-1)](https://ailandscape.org/tool/cognition-swe-1): Creator of SWE-1.5 — a fast-agent frontier model with near-SOTA code performance
- [Composer](https://ailandscape.org/tool/composer): Cursor (Oct 2025) — 4x faster than similarly capable models; built for agents
- [Devstral 2](https://ailandscape.org/tool/devstral-2): Mistral AI (Dec 2025) — open-source SOTA for code agents; 123B and 24B sizes
- [Gemini 3](https://ailandscape.org/tool/gemini-3): Google (Nov 2025) — beats Gemini 2.5 Pro at coding; masters agentic workflows
- [GLM](https://ailandscape.org/tool/glm): Z.ai (Jan 2026) — local coding and agentic assistant; the new 30B class standard
- [Codex](https://ailandscape.org/tool/codex): OpenAI (Feb 2026) — top agentic coding model; 25% faster, helped create itself

---

Part of [AI Landscape](https://ailandscape.org), an open map of the AI ecosystem. Web page: https://ailandscape.org/tool/code-world-model · Index for AI assistants: https://ailandscape.org/llms.txt