Training a Smash Bros World Model on a Budget
KrishRShah · x · 2026-07-20
An indie developer argues that world models aren't truly open-source if only massive labs can afford them. To prove this, he successfully trained a world model for Super Smash Bros. (Smash Melee) for just $250.
Key highlights of the project:
- Ultra-low training cost: 200x cheaper than conventional methods, with meaningful ablation studies costing under $5.
- Data & Open Source: Utilized 500 hours of action-aligned gameplay data. Code, data, model weights, and W&B experiment logs are fully open-sourced.
This demonstrates that everyday developers can engage in world model research at a fraction of the typical cost.
More from Research
- Token Collapse: Study Finds Graph Soft Prompting Tokens in LLMs Fail to Encode Structure — Sapienza · 2026-10-27
- Mathathon 2.0 launches with Terence Tao's SAIR, $1.5M in compute for teams — _sathvikr · 2026-10-27
- The 'sweet lesson' is real: two kinds of ML problems need opposite strategies — mike64_t · 2026-10-27
- PixGR: Pixel-space generative decoding for fast feed-forward 3D reconstruction — kwangmoo_yi · 2026-10-27
- UniAction: unified action conditioning lets one world model handle mixed action types — ZeYanjie · 2026-10-27
- Infini-AI Lab open-sources Hermit, a framework for agent harness evolution and continual learning — BeidiChen · 2026-10-27