SPADE: Self-Play in Adaptive Synthetic Executable Environments
Continuous self-improvement requires an ever-expanding pool of self-generated, diverse, adaptive goals. For language agents, existing training environment pools (hand-curated, statically synthesized, or frozen-verifier) keep the goal distribution fixed as the learner scales. We introduce SPADE (Self-Play in Adaptive Synthetic Executable Environments), a self-play RL framework in which a single LLM plays two roles: an Environment Designer that writes complete, long-horizon training environments as executable code with an OpenAI Gym-style reset()/step() interface, and a Reasoning Agent that learns to act in them. Each is a stateful, multi-turn environment (state transitions, reward function...
| Authors | Bo Liu, Simon Yu, Yiding Jiang et al. |
|---|---|
| Primary category | cs.CL |
| Published | 19 August 2026 |
| Last updated at source | 31 August 2026 |
Connected in the knowledge graph
- spade-rl/SPADE-Qwen3-30B-A3B-GamesModel · cited by
- spade-rl/SPADE-Environments-Qwen3-30B-GamesDataset · cited by
- spade-rl/SPADE-Qwen3-8B-GamesModel · cited by
- spade-rl/SPADE-Environment-Pool-GPT5.5-GamesDataset · cited by
- spade-rl/SPADE-Qwen3-4B-GamesModel · cited by
- spade-rl/SPADE-Environments-ToolUseDataset · cited by
Models and datasets referencing this paper
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.