← Back to feed

SlopCodeBench: Benchmarking How Coding Agents Degrade over Long-Horizon Tasks

AI · Studies · 2026-11-14 15:11

SlopCodeBench is a new benchmark designed to quantify the performance degradation of AI coding agents when handling long-horizon, complex tasks. It provides a standardized method for measuring how agent capabilities decline over extended workflows, highlighting critical limitations in current automated coding systems.

Source: arxiv.org


Auto-generated summary for reference only. RSS