FlashRSI open-sources RSI research: 12 algorithms, 30+ benchmarks, OOD gains still elusive
HuaxiuYaoML · x · 2026-11-01
FlashRSI is a new open-source toolkit for Recursive Self-Improvement (RSI) research, letting you run [Algorithm × Bench × Harness] evolution in one line of code: 12 RSI algorithms, 30+ benchmarks/datasets, and 3 target harnesses including Mini-Agent and Claude Code.
The companion RRSI paper (arXiv:2609.24972, work done during a Google internship) adds regularization to harness self-evolution to curb overfitting to training tasks. Key findings from the benchmark grid across 11 benchmarks and two harnesses:
- RRSI ranks #1/#2 across most benchmarks, with mean gains of +9.09 (Claude Code) and +6.07 (Mini-Agent);
- Most RSI methods show strong in-domain/near-domain gains that shrink or vanish on OOD benchmarks;
- Restricting evolution to skills/prompts alone can sometimes work even better.
More from coding & agent
- Agent Harness engineering resources: Microsoft Agent Framework, Anthropic evals, papers — goyalshaliniuk · 2026-11-01
- An Agent Saying 'Task Completed' Isn't Proof It Was Done Right — goyalshaliniuk · 2026-11-01
- Least Privilege and Checkpoints: Two Pillars of Reliable AI Agents — goyalshaliniuk · 2026-11-01
- Build a Bounded Control Loop to Keep AI Agents From Spinning — goyalshaliniuk · 2026-11-01
- Dimillian's Codex goes off the rails, pulls up Backrooms images as game references — Dimillian · 2026-11-01
- Dev proposes cross-device agent session mesh over tailnet, tags OpenAI and Anthropic — MannyKayy · 2026-11-01