AI Paper Atlas

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

Open-World LLM Logical Reasoning

Ye Mo, Chuan Zhou, Fengxiang Cheng, Jialin Yu, Liangming Pan, Fenrong Liu, Sheng Zhou, Haoxuan Li, Zhouchen Lin, Philip Torr

ICML 2026 · Large language models › Agents, code and math · approach: LLM planning with symbolic tools

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 translator and logic verifier to a CoT baseline (ablation)OpenIKLR · Multi-LogiEval 90%F1: 58.78 -> 61.49Improved
add fact verifier (ablation)OpenIKLR · Multi-LogiEval 90%F1: 73.25 -> 78.39Improved
add the missing-premise completion module (ablation)OpenIKLR · Multi-LogiEval 90%F1: 66.39 -> 73.25 across table rowsImproved
prefer a minimal premise set in the completion prompt (ablation)OpenIKLR · Multi-LogiEval 90%accuracy: 64.52 -> 68.53 (DeepSeek-V3.2); avg premises 6.18 -> 1.78Improved
minimal instead of maximal premise constraint (ablation)OpenIKLR · FOLIO 90%token cost: 3572.94 -> 3106.92 tokens; accuracy 85.28 -> 90.81Improved
add RAG retrieval (HyDE or SiReRAG) before OpenIKLR (ablation)OpenIKLR · Multi-LogiEval 90%accuracy: 68.53 -> 59.71 (HyDE, DeepSeek-V3.2)Hurt
increase max completion iterations Kmax (ablation)OpenIKLR · Multi-LogiEval 90%F1Improved
apply OpenIKLR to a small modelQwen3-8B · Multi-LogiEval 90%accuracy: 47.78 -> 66.85Improved
use symbolic-solver baselines (LINC, LogicLM) in open-world settingDeepSeek-V3.2 · Multi-LogiEval 90%accuracy: 50.64 (CoT) -> 20.74 (LINC)Hurt
apply OpenIKLR on legal QADeepSeek-V3.2 · LegalBench SARAF1: 50.21 -> 58.17Improved

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