SPADE adaptive self-play boosts 30B agents by 5.3 points over fixed RL
_AndrewZhao · x · 2026-08-22
SPADE (Self-Play in Adaptive Synthetic Executable Environments) is a new self-play RL framework that addresses the saturation issue in agent training caused by fixed task pools.
Core Mechanism:
- A single LLM plays two roles: Environment Designer and Reasoning Agent.
- The Designer writes complete, executable multi-turn environment code (OpenAI Gym-style reset()/step() interface).
- The Agent learns to act within these environments. The system uses hint-based regret to optimize, allowing the Designer to continuously generate tasks right at the Agent's capability frontier.
Results:
- At 30B scale, SPADE outperforms the strongest fixed-environment baseline by an average of +5.3 points across 8 held-out benchmarks (math, science, code, reasoning).
- Achieves significant gains in tool use: +5.7 on BFCL-v4 multi-turn and +13.9 on ACEBench-A.
Related event: SPADE: Single-Model Self-Play Writes Its Own Training Environments(13 posts)→
More from coding & agent
- Meta open-sources Muse Spark Open agent model with day-0 TPU support in SGLang — BanghuaZ · 2026-10-16
- Saying "load-bearing" in your prompt makes Opus 5.5 stop holding back — zacharynado · 2026-10-16
- You.com partners with Paperclip to give every agent live web access — PolarBearby · 2026-10-16
- I priced every Grok Bot call: $20 to $200 plan in 5 days, long jobs eat 52% — jonathanmalkin · 2026-10-16
- GitHub is criminally underrated as a free data store, devs argue — blelbach · 2026-10-16
- Claude Desktop doesn't support MCP sampling — here's a workaround for two-way MCP apps — wuj · 2026-10-16