Tencent's Terminal-RSI doubles 27B model's terminal task success via recursive self-improvement

Tencent Hy Frontier, with NUS and University of Maryland, introduced Terminal-RSI, which alternates recursive self-rewriting and policy refinement to let a 27B Qwen3-based agent improve itself with zero human involvement, boosting terminal task success from 20.68% to 50.58% in two iterations.

2026-11-08 ~ 2026-11-09 · 4 related posts

• Tencent's Terminal-RSI: a 27B model that rewrites and retrains itself with zero humans — Scobleizer · 2026-11-08
• Terminal-RSI boosts 27B model's terminal-agent success from 20.7% to 50.6% via recursive self-improvement — Dr_Singularity · 2026-11-09
• Terminal-RSI: self-improving agent system lifts 27B model success rate from 20.68% — Dr_Singularity · 2026-11-09

1 near-duplicate retellings: mark_k