AI Paper Atlas

Research map / Multi-agent systems and games / Human-AI and goal-conditioned

WebSeer: Training Deeper Search Agents through Reinforcement Learning with Self-Reflection

Guanzhong He, Zhen Yang, Jinxin Liu, Bin Xu, Lei Hou, Juanzi Li

ICLR 2026 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Language-grounded reinforcement agents · 7 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 self-reflective multi-answer submission during RL (vs single answer only) (ablation)WebSeer · MHQA benchmarksaccuracyImproved
add cold-start SFT initialization before RL (ablation)WebSeer · MHQA benchmarksaccuracyImproved
apply SFT to 3B modelQwen2.5-3B · HotpotQAaccuracy: drops 3.52 pointsHurt
apply SFT to 3B modelQwen2.5-3B · SimpleQAaccuracy: gains 8.06Improved
apply SFT to 7B modelQwen2.5-7B · HotpotQA and SimpleQAaccuracyHurt
scale to 14B model for SFT then RLWebSeer · HotpotQA and SimpleQAaccuracyImproved
increase share of long multi-refinement trajectories in SFT data (ablation)WebSeer SFT · HotpotQAtool calls and accuracyMixed
use web search at inference instead of local RAGWebSeer · 7 MHQA benchmarks in-domain avgaccuracy: 78.9 -> 82.6Improved
WebSeer vs Search-r1 in-domainWebSeer · In-domain avgaccuracy: exceeding Search-r1 by 12.5 pointsImproved
WebSeer vs prior best on Bamboogle (local RAG)WebSeer · Bamboogleaccuracy: 81.6, 12.8-point improvementImproved
add RL after SFT for 3B and 7B modelssmall models · HotpotQAaccuracyNo clear effect

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