Crusoe is the exclusive compute partner for General Reasoning's RLCR agent-organization research
NewsX
General Reasoning introduced Reinforcement Learning from Collective Rewards, a recipe that trains language models to work as an organization of agents sharing a message board and workspace, and used it to train a 120B model called Avalon-M1-Mini. Crusoe said it was the exclusive compute partner for the launch. The research announcement drew strong attention on X, with close to two thousand likes and nearly two thousand bookmarks. It is one of several recent model releases, alongside Gradium's on-device Phonon speech models, that Crusoe has highlighted as trained on its cloud.