Code World Model: A Coding Agent as World Brain, Code for Rules, Video for Rendering
burny_tech · x · 2026-08-28
The paper Code World Model: Coding Agent as World Brain argues that world models should keep reasoning and causality in code rather than learning dynamics entirely from pixels.
- A coding agent serves as the world brain, reasoning about events and generating executable code that maintains persistent world state and rule-consistent evolution.
- A lightweight proxy representation (frame-wise spatiotemporal constraints compiled into a proxy video) bridges executable state and video generation—code decides what happens, the video model decides how it looks.
- Data pipelines build aligned proxy-observation pairs from gameplay and real-world videos.
- After fine-tuning on paired gameplay data, MiniMax-H3 follows spatiotemporal specs from simple interactive worlds built by the coding agent while preserving rich visual detail and dynamics.
The authors see this as a new path toward open-ended world models.
Related event: Code World Model: Coding Agent as the Brain of World Models(2 posts)→
More from coding & agent
- Flova Advances to Agent-Native Video Creation with All-in-One Tools — SimplyAnnisa · 2026-08-28
- Anthropic tests Hub Mode in Claude for sub-agent management — testingcatalog · 2026-08-28
- Anthropic tests 'Hub' feature for task orchestration in Claude — testingcatalog · 2026-08-28
- Weak RL may trigger communication traits hidden in pretraining — xuanalogue · 2026-08-28
- Tool call dispositions may generalize to unsanctioned agent communication — xuanalogue · 2026-08-28
- OpenAI agent message-leaving may emerge from pretraining priors — xuanalogue · 2026-08-28