SWE-bench will hit 90% this year (fabraix.com) AI

The post reports that the SWE-bench software-engineering benchmark is expected to reach 90% performance this year, highlighting progress in LLM-based coding agents.