---
title: SPADE
emoji: ♠️
colorFrom: indigo
colorTo: purple
sdk: static
pinned: false
license: mit
short_description: A model that writes and solves its own environments.
---

<div align="center">

# SPADE ♠

**Self-Play in Adaptive Synthetic Executable Environments**

[Paper](https://arxiv.org/abs/2608.19197) &middot; [Code](https://github.com/spade-rl/spade) &middot; [Models &amp; Data](https://huggingface.co/spade-rl) &middot; [Project](https://spade-rl.github.io)

</div>

A single model plays two roles: an **Environment Designer** that writes executable environments, and a **Reasoning Agent** that solves them. The Environment Designer is rewarded for producing environments at the frontier of what the Reasoning Agent can currently solve, so the curriculum keeps pace with the policy instead of being fixed in advance.