Tsinghua's Frontis-MA1 pushes 35B model to 71.21% on MLE-Bench Lite for recursive self-improvement

ceciletamura · x · 2026-09-03

A Tsinghua University team released Frontis-MA1, a paper studying recursive self-improvement (RSI) using machine learning engineering (MLE) as an executable testbed for AI4AI.

Original post →

More from coding & agent

coding & agent channel →