Xiaomi's CodeMidas turns open-source code into 5,545 RL environments, boosting coding agents up to 17%
eliebakouch · x · 2026-09-21
Xiaomi MiMo's CodeMidas: Scaling Coding RL Environments from Source Code
Xiaomi's MiMo team presents CodeMidas, an agentic pipeline that converts implemented functionality in existing codebases into executable RL environments, using source code alone—no reliance on issues or commits.
Agentic pipeline highlights:
- Agents explore code to write behavioral specifications and build execution-grounded tests
- Tasks are validated via execution checks and repeated solution rollouts
- Robustness is tested by letting agents try to cheat; difficulty measured via pass@n
Results:
- 5,545 training tasks from 3,185 open-source repos, spanning 23 languages and 15 domains
- Training MiMo-V2.5 with GRPO improves all five benchmarks: DeepSWE +11.7%, ProgramBench +17%, Terminal-Bench v2.1 +8.5%
- Ablations show more high-quality tasks is the key scaling axis
Related event: Xiaomi's CodeMidas Auto-Generates 5,545 RL Environments from Source Code(4 posts)→
More from coding & agent
- Stripe's Emily Sands on agentic commerce: compute is the new cash, fraud is now full-funnel — every · 2026-09-22
- LangChain offers open-source decision model SemIf free for a week in LangSmith Gateway — hwchase17 · 2026-09-22
- Writing skills for models and their harnesses is one of the highest-leverage skills now — pvncher · 2026-09-22
- Grok VS Code extension now listed on Open VSX and the VS Code Marketplace — PawelHuryn · 2026-09-22
- LLMs Are Too Big: Aronchick on Right-Sizing Models for Log Routing — aronchick · 2026-09-22
- Dev syntax-highlights nested string literals inside a coding agent, making prompts readable again — lucasmeijer · 2026-09-22