AI Paper Atlas

Research map / Large language models / Alignment and preferences

SWE-IF: Aligning Code Evaluation with Human Preference

Ming Zhong, Xiang Zhou, Ting-Yun Chang, Qingze Wang, Nan Xu, Xiance Si, Dan Garrette, Shyam Upadhyay, Jeremiah Zhe Liu, Jiawei Han, Benoit Schillings, Jiao Sun

ICML 2026 · Large language models › Alignment and preferences · approach: LLM-as-judge and feedback data · 0 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 more non-functional instructions (1 to 5), single-turnall evaluated LLMs · Big-SWE-IFfunctional regression rate (%): 2.48 -> 5.76Hurt
add more non-functional instructions (1 to 5), multi-turnall evaluated LLMs · Big-SWE-IFfunctional regression rate (%): 3.18 -> 9.31Hurt
use multi-turn editing instead of single-turn generationall evaluated LLMs · Big-SWE-IF / Live-SWE-IFfunctional regression: 5.76 -> 9.31 (5 instructions, Big-SWE-IF)Hurt
use multi-turn editing instead of single-turn generationall evaluated LLMs · Big-SWE-IFtask-level IF score: 3% to 4.5% advantageImproved
use multi-turn editing instead of single-turn generationall evaluated LLMs · Live-SWE-IFtask-level IF score: around 8% advantageImproved
add more instructions (1 to 5)Claude 4 Opus, single-turn · Big-SWE-IFtask-level IF score: 88.77 -> 46.75Hurt
add more instructions (1 to 5)Kimi K2, single-turn · Live-SWE-IFtask-level IF score: 62.75 -> 11.94Hurt
add more instructions (1 to 5)o4 mini, multi-turn · Live-SWE-IFfunctional regression rate (%): 3.63 -> 15.92Hurt
add more instructions (1 to 5)Claude 4 Opus, single-turn · Big-SWE-IFfunctional regression rate (%): -0.86 -> -2.08No clear effect
mix IF with functional score (alpha=0.4) rather than Func aloneLMArena Elo correlation · Big-SWE-IFPearson correlation with LMArena Elo: alpha=0.4 optimumImproved
mix IF with functional score (alpha=0.7) rather than either aloneLMArena Elo correlation · Big-SWE-IFSpearman correlation with LMArena Elo: alpha=0.7 optimumImproved
use pure IF instead of pure FuncLMArena Elo correlation · Big-SWE-IFSpearman correlation with LMArena Elo: over 0.1 points higherImproved

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