AI Paper Atlas

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

Beyond ReAct: A Planner-Centric Framework for Complex Tool-Augmented LLM Reasoning

Xiaolong Wei, Yuehu Dong, Xingliang Wang, Xingyu Zhang, Zhejun Zhao, Dongdong Shen, Long Xia, Dawei Yin

AAAI 2026 · Large language models › Agents, code and math · approach: LLMs for optimization and heuristics · 23 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
add GRPO RL stage after SFT (ablation)Qwen3-8B planner · ComplexTool-Plan EasyDAG exact match: 0.781 -> 0.803Improved
add GRPO RL stage after SFT (ablation)Qwen3-8B planner · ComplexTool-Plan HardDAG exact match: 0.295 -> 0.319Improved
add GRPO RL stage after SFT (ablation)Qwen3-4B planner · ComplexTool-Plan HardDAG exact match: 0.241 -> 0.259Improved
add GRPO RL stage after SFT (ablation)Qwen3-1.7B planner · ComplexTool-Plan EasyDAG exact match: 0.723 -> 0.756Improved
add GRPO RL stage after SFT (ablation)Qwen3-4B planner · ComplexTool-Plan Hardedge F1: 0.627 -> 0.621Hurt
add GRPO RL stage after SFT (ablation)Qwen3-8B planner · ComplexTool-Plan Hardedge F1: 0.657 -> 0.659No clear effect
scale planner size from 1.7B to 8B (SFT+RL) (ablation)Qwen3 planner · ComplexTool-Plan HardDAG exact match: 0.218 -> 0.259 -> 0.319Improved
use fine-tuned Qwen3-8B planner instead of prompted proprietary LLMDAG planning · ComplexTool-Plan HardDAG exact match: 0.106 -> 0.319Improved
use plan-then-execute planner (Qwen3-8B RL) instead of ReAct with GPT-4StableToolBench · StableToolBench averageSoPR: 48.2 -> 59.8Improved
use Qwen3-8B planner instead of GPT-4 (DFSDT)StableToolBench · StableToolBench averageSoPR: 70.3 -> 59.8Hurt
use global DAG planning instead of iterative DTA-LlamaStableToolBench · StableToolBenchinference steps: 2.48 -> 2.29Improved

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