Digital Agent outperforms o1 by 15% – trained with new RL-variant similar to R1 (arxiv.org) 11 pts|1 year ago|discuss
