U.S. open-weight model lab targeting enterprises and developers who need portable, sovereign AI. Their Trinity family spans sparse MoE models from a 6B-parameter Nano to a 400B-parameter Large, all pre-trained from scratch on up to 17 trillion tokens, using the Muon optimizer and a proprietary load-balancing technique called SMEBU, released under Apache 2.0.
Loka post-trained Arcee's open 26B Trinity Mini with RL to run multi-step biomedical research workflows, hitting 81.2% on drug-tool evaluation and 0.863 on biological reasoning
Nathan Lambert, the researcher behind Ai2's OLMo and Tulu and author of the RLHF Book, joins Arcee as Research Advisor as the open-source AI movement faces its most critical inflection point yet.