AI Paper Atlas

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

RefineBench: Evaluating Refinement Capability in Language Models

Young-Jun Lee, Seungone Kim, Byung-Kwan Lee, Minkyeong Moon, Yechan Hwang, Jong Myoung Kim, Graham Neubig, Sean Welleck, Ho-Jin Choi

ICLR 2026 · Large language models › Agents, code and math · approach: Mathematical and structural reasoning benchmarks

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 full checklist as evaluation criteria to self-refinement (ablation)Gemini 2.5 Pro · RefineBenchPass@5: 31.3 -> 75.8Improved
add full checklist as evaluation criteria to self-refinement (ablation)LLaMA-3.1-70B-Instruct · RefineBenchPass@5: 4.6 -> 48.2Improved
add explicit checklist feedback (guided refinement) instead of self-refinementClaude-Opus-4.1 · RefineBenchPass@5: 20.8 -> 98.4Improved
add explicit checklist feedback (guided refinement)LLaMA-3.1-8B-Instruct · RefineBenchPass@5: +28.7Improved
use reasoning (thinking) variant instead of instructQwen3-30B-A3B · RefineBench self-refinementavg improvement: -1.6 -> +1.4Improved
use reasoning model o1 instead of GPT-4oself-refinement · RefineBenchavg improvement: -1.4 -> -0.2Improved
more self-refinement turns (5 vs 1)Claude-Sonnet-4 · RefineBenchPass@t: +0.8No clear effect
more self-refinement turns (5 vs 1)DeepSeek-R1 · RefineBenchPass@t: -0.1No clear effect
more self-refinement turns (5 vs 1)DeepSeek-R1 (Qwen-32B) · RefineBenchPass@t: -2.5Hurt
provide only 50% of checklist feedback (partial guidance)LMs · RefineBenchPass@tMixed
longer termination turn (continue refining more turns)reasoning LMs · RefineBenchPass@5 correlation: p-value < 0.01, R2 = -0.477Hurt
more reasoning tokens for self-refinementGemini 2.5 Pro · RefineBenchPass@tNo 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