Arcee AI

Arcee AI is a San Francisco lab building frontier open-weight foundation models under the Apache 2.0 licence. Founded in 2023 by Mark McQuade — an early employee at Hugging Face — with Jacob Solawetz and Brian Benedict, it started as a post-training and customization shop, taking open Llama, Mistral and Qwen models and fine-tuning them for enterprise clients such as SK Telecom. In 2025 McQuade bet the company on training models of its own: over roughly six months, on 2,048 Nvidia Blackwell B300 GPUs, Arcee trained the Trinity family for about $20M all-in — 65-70% of the capital it had at the time. Trinity Nano (6B) and Trinity Mini (26B) shipped in December 2025, followed in January 2026 by Trinity Large, a 400B-parameter sparse mixture-of-experts model with 13B active parameters per token that Arcee benchmarks against Meta's Llama 4 Maverick and Z.ai's GLM-4.5 and calls among the largest open-source foundation models trained and released by a U.S. company. On 2026-09-16 Arcee announced a Series B led by Vista Equity Partners, Cambium Capital and Emergence Capital valuing it at more than $1B — Fortune reported a $1B pre-money and put the raise at at least $150M, though the company disclosed no amount. Proceeds fund the next generation of Trinity models, an expanded collaboration with the U.S. Department of Energy and its 17 national laboratories on Genesis-Science-1, and products for customizing, evaluating and deploying open models. Roughly 30 employees as of early 2026; the last public revenue figure is ~$2M from mid-2024.