Arcee Trinity Review — 30-Person Startup Builds 400B Open-Source LLM for $20M

Arcee AI — a 30-person San Francisco startup — trained Trinity Large, a 400B-parameter sparse MoE model, in 33 days on $20M of compute, then released it under Apache 2.0. We review the Trinity family (Nano, Mini, Large, Large Thinking), benchmark numbers, pricing, the enterprise self-hosting case, and what a startup can actually achieve against frontier labs in 2026.