Arcee AI pivoted from post-training to building models from scratch. It spent $20 million to train four open-weight models. Its largest model, Trinity Large, has 400 billion parameters. Trinity Large outperformed Meta’s Llama 3 and matched Mistral and Chinese models. Arcee raised a Series B at a $1 billion pre-money valuation. Vista Equity Partners, Cambium Capital, and Emergence Capital led the round. Microsoft’s M12, Hitachi, and Wipro also participated. The funding will go toward new models, products, and a U.S. Department of Energy partnership. Arcee aims to catch up to leading Chinese open-weight models. Its founder bet 70% of its capital on the pivot.

• $20 million to train four models
• $1 billion pre-money valuation
• at least $150 million raised in Series B
• 400 billion parameters for Trinity Large