AI Paper Atlas

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

Local Success Does Not Compose: Benchmarking Large Language Models for Compositional Formal Verification

Xu Xu, Xin Li, Xingwei Qu, Jie Fu, Binhang Yuan

ICLR 2026 · Large language models › Agents, code and math · approach: Code generation 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
sample more attempts, Pass@1 to Pass@8all 13 LLMs · DafnyCompverified rate: average gain from Pass@4 to Pass@8 only 0.8%No clear effect
sample more attempts, Pass@1 to Pass@8all 13 LLMs · DafnyCompsyntax correct rate: +7.3% on average (Pass@4 to Pass@8)Improved
sample more attempts, Pass@1 to Pass@8claude-3.5-sonnet · DafnyCompverified rate: 3.67 -> 7.00Improved
use reasoning-specialized models (QwQ-32B, DeepSeek-R1)LLMs · DafnyCompverified rate: QwQ-32B 0.00 at Pass@8; DeepSeek-R1 0.33No clear effect
scale model size (480B vs 32B)Qwen coder models · DafnyCompverified rate: 1.00 vs 0.67 at Pass@8No clear effect
move from single-function (DafnyBench) to multi-function (DafnyComp) tasksLLMs · Dafny specification generationverification success: ~53% -> 3.69%Hurt
use code-specialized instead of general-purpose modelsLLMs · DafnyCompverified rateNo clear effect
use chain-based composition instead of arbitrary call graphs (ablation)benchmark synthesis · synthesis successsuccess rate: <8% -> 47%Improved
use incremental AST-guided translation with verifier refinement instead of direct Python-to-Dafny translation (ablation)benchmark synthesis · Python to Dafny translationsuccess rate: <5% -> 47%Improved

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