RExBench: Can coding agents autonomously implement AI research extensions?

TL;DR

RExBench is a new benchmark designed to evaluate whether large language model agents can autonomously extend research papers, revealing current limitations in their ability to handle complex research tasks without human guidance.

Contribution

The paper introduces RExBench, a benchmark for assessing LLM agents' capability to autonomously perform research extensions, and evaluates existing agents showing significant room for improvement.

Findings

All agents failed to autonomously implement most research extensions.

The best agent achieved only about 33% success rate without human hints.

Success rate improved to below 44% with additional human guidance.

Abstract

Agents based on Large Language Models (LLMs) have shown promise for performing sophisticated software engineering tasks autonomously. In addition, there has been progress towards developing agents that can perform parts of the research pipeline in machine learning and the natural sciences. We argue that research extension and its implementation is a critical capability for such systems, and introduce RExBench to support the evaluation of this capability. RExBench is a benchmark consisting of realistic extensions of 12 research papers that aim to investigate novel research hypotheses. Each task is set up as an extension to an existing research paper and codebase, accompanied by domain expert-written instructions. RExBench is robust to data contamination and supports an automatic evaluation infrastructure that executes agent outputs to determine whether the success criteria are met. We use this benchmark to evaluate 12 LLM agents implemented using two different frameworks, aider and OpenHands. We find that all agents fail to autonomously implement the majority of the extensions, with the best agent achieving around a 33% success rate. Although the success rate improves with additional human-written hints, the best performance under this setting remains below 44%. This indicates that current agents are still short of being able to handle realistic research extension tasks without substantial human guidance.