← 返回归档

帖子详情

查看 X 原文
O
elvis@omarsar0
Banger paper from Tencent on recursive self-improvement for terminal agents. (bookmark it) This is a great read if you are tracking research on training terminal or coding agents with RL Starting from Qwen3.8-27B, they alternate two phases over five stages. One phase fine-tunes on rewritten versions of trajectories that passed the verifier. The other runs PPO with a critic inherited from the previous stage. The four-suite Terminal-Bench average goes from 20.68% to 50.58%, with 86.52% on Terminal-Bench 2.1 and 79.00% on Terminal-Bench Hard. The reward audit is just as useful. Under a dense progress reward, failed runs collected 50.09% of the reward, and 48 failures passed every executed check but still missed the final test. Separating strict success from partial credit cut that share to 28.72%. arxiv.org/abs/2610.22192 Chat with Paper: academy.dair.ai/papers/termina…
媒体保存失败：[Errno 104] Connection reset by peer
帖子 ID
2111487480281416004
Markdown
/app/data/archive/2026/10/17/omarsar0/2111487480281416004/post.md
保存状态
complete
查看 Markdown 内容
---
tweet_id: "2111487480281416004"
author: "omarsar0"
published_at: "2026-10-17T16:00:19+00:00"
source_url: "https://x.com/omarsar0/status/2111487480281416004"
---

Banger paper from Tencent on recursive self-improvement for terminal agents.  
  
(bookmark it)  
  
This is a great read if you are tracking research on training terminal or coding agents with RL  
  
Starting from Qwen3.8-27B, they alternate two phases over five stages. One phase fine-tunes on rewritten versions of trajectories that passed the verifier. The other runs PPO with a critic inherited from the previous stage.  
  
The four-suite Terminal-Bench average goes from 20.68% to 50.58%, with 86.52% on Terminal-Bench 2.1 and 79.00% on Terminal-Bench Hard.  
  
The reward audit is just as useful. Under a dense progress reward, failed runs collected 50.09% of the reward, and 48 failures passed every executed check but still missed the final test. Separating strict success from partial credit cut that share to 28.72%.  
  
[arxiv.org/abs/2610.22192](https://arxiv.org/abs/2610.22192)  
  
Chat with Paper: [academy.dair.ai/papers/termina…](https://academy.dair.ai/papers/terminal-rsi-recursive-self-improving-terminal-agents-via-recursive-self-rewrite-2610.22192)

![Tweet Image](https://pbs.twimg.com/media/HU2CrEUbAAAvNWE.jpg)