Selene Blog

Long-form notes on building local-first multi-agent products

Borrowing modern reading UX patterns from 2025-2026 blogs: clear cards, useful metadata, side navigation, and media-rich storytelling with practical implementation detail.

SWE-bench Lite CLI result showing 182 resolved out of 300

Mar 14, 20266 min read

My first SWE-bench Lite run with Selene cleared 60.67%

This was my first real SWE-bench Lite pass with Selene, not a polished rerun. I used Claude Opus 4.6 in non-thinking mode, kept the default Selene agent, ran tasks sequentially, and still landed at 182 resolved out of 300.

EngineeringBenchmarksSWE-bench Litebenchmark