Papers with Code adds parameter-size filtering, exposing sub-6B SWE-Bench Verified SOTA

NielsRogge · x · 2026-09-09

NielsRogge shares that Papers with Code now lets users filter benchmark entries by parameter size. A notable example: SWE-Bench Verified, the 500-sample human-validated subset testing real-world software issue solving, can now be filtered to show SOTA only among models with ≤6B parameters — handy for tracking small-model progress. The board aggregates evals from a wide range of papers spanning Claude, DeepSeek, Gemini, Kimi, MiniMax and more.

Original post →

More from coding & agent

coding & agent channel →