AI Paper Atlas

Research map / Large language models / Agents, code and math

Think Fast and Slow: Step-Level Cognitive Depth Adaptation for LLM Agents

Ruihan Yang, Fanghua Ye, Xiang Wei, Ruoqing Zhao, Kang Luo, Xinbo Xu, Bo Zhao, Ruotian Ma, Shanyi Wang, Zhaopeng Tu, Xiaolong Li, Deqing Yang and 1 more

ICML 2026 · Large language models › Agents, code and math · approach: Agent instruction-following benchmarks · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use average log-probability instead of min log-probability as confidence (ablation)CoPO Qwen2.5-7B · ALFWorldsuccess rate: 81.5 -> 92.5Improved
use average log-probability instead of entropy as confidence (ablation)CoPO Qwen2.5-7B · ALFWorldsuccess rate: 79.0 -> 92.5Improved
use average log-probability instead of max log-probability as confidence (ablation)CoPO Qwen2.5-7B · ALFWorldsuccess rate: 89.0 -> 92.5Improved
add CoSFT cold-start initialization before RL (ablation)CoPO Qwen2.5-7B · ALFWorldsuccess rate: 86.0 -> 92.5Improved
add balanced CoSFT initialization instead of expert-selected levels (ablation)CoPO Qwen2.5-7B · ALFWorldsuccess rate: 87.5 -> 92.5Improved
reweight only successful trajectories instead of both (ablation)CoPO Qwen2.5-7B · ALFWorldsuccess rate: 83.0 -> 92.5Improved
add cold start (ablation)CoPO Qwen2.5-7B · ScienceWorldsuccess rate: 64.5 -> 72.0Improved
replace GRPO with CoPOCoSFT Qwen2.5-7B · ALFWorld+ScienceWorld averagesuccess rate: 68.3 -> 82.3Improved
replace GiGPO with CoPOCoSFT Llama3.1-8B · ALFWorld+ScienceWorld averagesuccess rate: 76.8 -> 81.0Improved
CoPO instead of GRPO token usageCoSFT Qwen2.5-7B · ALFWorld+ScienceWorld averagetokens per trajectory: 4367.3 -> 1641.4Improved
CoSFT_exp instead of SFTQwen2.5-7B · ALFWorld+ScienceWorld averagesuccess rate: 40.8 -> 54.0; tokens 577.5 -> 3562.4Mixed
R1-distilled variant instead of vanillaQwen2.5-7B ReAct · ALFWorld+ScienceWorld averagesuccess rate: 26.5 -> 2.5Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy