SWE-bench Multimodal: Do AI Systems Generalize to Visual Software Domains?

TL;DR

This paper introduces SWE-bench Multimodal, a new benchmark for evaluating AI systems' ability to fix bugs in visual, JavaScript-based software, revealing current limitations and demonstrating the effectiveness of a language-agnostic approach.

Contribution

The paper presents SWE-bench Multimodal, a novel benchmark with visual tasks in JavaScript, and shows that a language-agnostic system outperforms existing models on this new domain.

Findings

Top systems struggle with visual, cross-language tasks.

SWE-agent resolves twice as many tasks as the next best system.

Visual and cross-language generalization remains a challenge for current AI systems.

Abstract

Autonomous systems for software engineering are now capable of fixing bugs and developing features. These systems are commonly evaluated on SWE-bench (Jimenez et al., 2024a), which assesses their ability to solve software issues from GitHub repositories. However, SWE-bench uses only Python repositories, with problem statements presented predominantly as text and lacking visual elements such as images. This limited coverage motivates our inquiry into how existing systems might perform on unrepresented software engineering domains (e.g., front-end, game development, DevOps), which use different programming languages and paradigms. Therefore, we propose SWE-bench Multimodal (SWE-bench M), to evaluate systems on their ability to fix bugs in visual, user-facing JavaScript software. SWE-bench M features 617 task instances collected from 17 JavaScript libraries used for web interface design, diagramming, data visualization, syntax highlighting, and interactive mapping. Each SWE-bench M task instance contains at least one image in its problem statement or unit tests. Our analysis finds that top-performing SWE-bench systems struggle with SWE-bench M, revealing limitations in visual problem-solving and cross-language generalization. Lastly, we show that SWE-agent's flexible language-agnostic features enable it to substantially outperform alternatives on SWE-bench M, resolving 12% of task instances compared to 6% for the next best system.