🛰️ Daily AI Frontier
‹ back to 2026-07-22

CodeRescue: Budget-Calibrated Recovery Routing for Coding Agents

Research LLM Agents

Merged summary

TL;DR - CodeRescue routes failed coding-agent attempts between additional cheap-model recovery and escalation to a stronger model. Its budget-calibrated router can adapt to changing cost constraints without retraining.

  • Trains a supervised router using execution feedback from coding-agent rollouts.
  • Uses Conformal Risk Control to select deployment-time cost penalties with marginal expected-cost control under exchangeability.
  • Cheap recovery and escalation show complementary success patterns across five coding benchmarks.
  • One GPT-5.4-nano/GPT-5.4 configuration surpassed always-escalate solve rate at 35% of its mean recovery cost.

Sources (1)

CodeRescue: Budget-Calibrated Recovery Routing for Coding Agents

arXiv cs.AI Qijia He, Jiayi Cheng, Chenqian Le, Rui Wang, Xunmei Liu, Yixian Chen, Jie Mei, Zhihao Wang, Xupeng Chen, Yuhuan Chen, Tao Wang 2026-07-21 arXiv:2607.19338

TL;DR - CodeRescue routes failed coding-agent attempts between additional cheap-model recovery and escalation to a stronger model. Its budget-calibrated router can adapt to changing cost constraints without retraining.

  • Trains a supervised router using execution feedback from coding-agent rollouts.
  • Uses Conformal Risk Control to select deployment-time cost penalties with marginal expected-cost control under exchangeability.
  • Cheap recovery and escalation show complementary success patterns across five coding benchmarks.
  • One GPT-5.4-nano/GPT-5.4 configuration surpassed always-escalate solve rate at 35% of its mean recovery cost.
item →