AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

In-The-Flow Agentic System Optimization for Effective Planning and Tool Use

Zhuofeng Li, Haoxiang Zhang, Seungju Han, Sheng Liu, Jianwen Xie, Yu Zhang, Yejin Choi, James Zou, Pan Lu

ICLR 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 61 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
train the planner on-policy with Flow-GRPO (vs frozen planner) (ablation)AgentFlow (Qwen2.5-7B) · 6-task average (Bamboogle, 2Wiki, GAIA, AIME24, AMC23, GameOf24)accuracy: 38.5 -> 55.7Improved
replace frozen Qwen2.5-7B planner with frozen GPT-4o (ablation)AgentFlow · 6-task averageaccuracy: 38.5 -> 44.3Improved
offline SFT on GPT-4o trajectories for the planner (ablation)AgentFlow · 6-task averageaccuracy: 38.5 -> 19.5Hurt
add Flow-GRPO to AgentFlow with 7B backboneAgentFlow · 2Wikiaccuracy: 60.0 -> 77.2Improved
add Flow-GRPO to AgentFlowAgentFlow · GameOf24accuracy: 31.0 -> 53.0Improved
add Flow-GRPO to AgentFlowAgentFlow · AMC23accuracy: 47.4 -> 61.5Improved
add Flow-GRPO to AgentFlowAgentFlow · GPQA / MedQAaccuracy: GPQA 37.0 -> 47.0; MedQA 76.0 -> 80.0Improved
apply Flow-GRPO fine-tuning to both 3B and 7B backbones (ablation)AgentFlow · multiple tasksaccuracyImproved
increase maximum turns at inference from 3 to 10 (ablation)AgentFlow (7B) · 2Wiki, GameOf24, AIME24, GAIAaccuracy: avg turns on GAIA 2.43 -> 5.42Improved
use Flow-GRPO-trained planner instead of untrained one (ablation)AgentFlow · MedQA tool useGoogle Search call ratio: 66.2% -> 10.9% Google Search; Wikipedia Search 0 -> 59.8%Improved
compare Flow-GRPO-trained AgentFlow with monolithic tool RL (ToRL)AgentFlow vs ToRL · AIME24 validation during trainingvalidation accuracyImproved

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