The round
Arcee AI, a US developer of open-weight foundation models, has raised a Series B round that values it at more than $1bn, the company announced on Wednesday. Vista Equity Partners, Cambium Capital and Emergence Capital led the round, with AI10 Ventures, Hitachi, IAG, M12, P7 and Wipro also taking part.
Arcee did not say how much it raised. Fortune, which reported the round first, said a source familiar with the matter put the amount at no less than $150m, and described the $1bn figure as a pre-money valuation. Fortune identified M12 as Microsoft’s venture fund.
What $20m bought
Arcee started out working on post-training, adapting existing models, according to Fortune. Less than a year ago it decided to build its own open-weight models from scratch, the company wrote in a blog post. Over six months it scaled from a 4.5-billion-parameter dense model to Trinity Large, which the release describes as a sparse mixture-of-experts model with 400 billion parameters, 13 billion of them active for each token.
Arcee says its entire 2025 model lineup, including Trinity Large, was built for approximately $20m, a figure it says covers salaries, compute, data, infrastructure and operations. Fortune framed the same number as the cost of training four models. Either way it is a company-reported figure, not an audited one, and far below what the largest labs spend.
“We saw an opportunity, and we had $30 million in the bank,” co-founder and chief executive Mark McQuade told Fortune, describing his decision to spend most of the company’s capital on building its own models.
Where the money goes
Arcee names three priorities. First, it will finish the next generation of Trinity, which it says is already in training, and build models at several sizes, from ones small enough for phones and laptops to larger systems for scientific and developer work.
Second, it will expand its work with the US Department of Energy and its national laboratories on Genesis-Science-1, an open model for scientific research. Third, it will build products that help organisations customise, evaluate, deploy and run open models.
In the release, Monti Saroya of Vista Equity Partners pointed to Arcee’s technical execution and capital efficiency as reasons for backing it. Lucas Atkins, the chief technology officer, said efficiency had shaped every part of how the company builds its models.
The open-weight race
Arcee presents Trinity Large as a permissively licensed model developed end to end in the United States, and says it was the first of its kind since Meta stopped releasing Llama models in early 2025. McQuade told Fortune that the US “kind of dropped the ball on open source”, and that Arcee is measuring itself against Chinese open-weight models, naming Z.ai’s GLM Flash as the target.
What to watch
Arcee has not given a release date for the next Trinity models, and it has not disclosed the size of the round or its post-money valuation.
