AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

G-reasoner: Foundation Models for Unified Reasoning over Graph-structured Knowledge

Linhao Luo, Zicheng Zhao, Junnan Liu, Zhangchi Qiu, Junnan Dong, Serge Panev, Chen Gong, Thuy-Trang Vu, Gholamreza Haffari, Dinh Phung, Alan Wee-Chung Liew, Shirui Pan

ICLR 2026 · Large language models › Retrieval and knowledge · approach: Knowledge graph reasoning with LLMs · 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 distillation loss from text encoder (ablation)G-reasoner · HotpotQAR@2: 81.1 -> 77.4 without itImproved
add distillation loss from text encoder (ablation)G-reasoner · 2WikiR@2: 75.6 vs 75.9 without itNo clear effect
add node text semantics (ablation)G-reasoner · HotpotQAR@2: 81.1 -> 79.4 without itImproved
add node text semantics (ablation)G-reasoner · 2WikiR@5: 96.1 -> 95.2 without itImproved
add graph foundation model (ablation)G-reasoner · HotpotQAR@2: 81.1 -> 11.6 without itImproved
add graph foundation model (ablation)G-reasoner · MuSiQueR@2: 52.1 -> 3.8 without itImproved
use G-reasoner as retriever on HippoRAG graph instead of personalized PageRankHippoRAG graph · HotpotQAEM: 46.3 -> 54.0Improved
use G-reasoner as retriever on LightRAG graph instead of embedding+graph searchLightRAG graph · HotpotQAEM: 9.9 -> 49.7Improved
use G-reasoner on Youtu-GraphRAG graph (unseen retriever, no fine-tuning)G-reasoner · MuSiQueEM: 30.3 vs 28.9 on HippoRAG graphImproved
BFloat16 mixed precision trainingG-reasoner training · A100 single GPUmemory and throughput: 80GB -> 66GB memory; 1.29 -> 2.72 samples/sImproved
replace multi-step RAG (Search-R1) with G-reasoner single passSearch-R1 · HotpotQAtime per sample (s): 0.496 -> 0.114Improved
use G-reasoner instead of HippoRAG on larger MuSiQue-Full corpusHippoRAG · MuSiQue-FullEM: 24.62 -> 33.64Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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