Recursive›Recursive› offers automated AI research system

Recursive's automated AI research system closes the research loop for a target objective by proposing ideas, implementing them, running experiments, validating results, and choosing the next experiment. It runs many parallel research threads, retains useful context, combines promising branches, and validates against reward hacks before accepting improvements.

The system has produced state-of-the-art results in fixed-budget language-model training, small-model training speed, and GPU kernel optimization, and the company has open-sourced artifacts from these runs.