Frontis-MA1: 35B Meta-Evolution Agent for Machine Learning Engineering
July 29, 2026
Frontis-MA1 is a 35B parameter model trained via execution-grounded SFT and RL to perform recursive self-improvement in MLE. It utilizes four program-evolution operators—Draft, Improve, Debug, and Crossover—to navigate long-horizon search tasks within the OpenMLE ecosystem.
HOW THIS AFFECTS YOU
●
builderYou can leverage these evolution operators to automate parts of the ML engineering lifecycle.
●
researcherThis provides a structured framework for studying recursive self-improvement in executable environments.