Arcee builds open-weight foundation models for developers and enterprise teams that want to inspect, adapt and run AI without sending every workload through a model vendor’s closed API. Fortune reported that the company has raised a Series B at a $1 billion pre-money valuation. Arcee declined to disclose the size, but a person familiar with the financing told Fortune it was at least $150 million.
Its Trinity family is released under the Apache 2.0 license. A developer can download the weights, inspect or fine-tune them and serve the resulting model in a public cloud, on premises or at the edge. Teams that do not want to operate that infrastructure can use Arcee’s hosted API. The models also integrate with agent frameworks and coding harnesses for applications that perform multistep work and call outside tools.
Trinity-Large-Thinking uses a sparse mixture-of-experts architecture with about 400 billion parameters overall and roughly 13 billion active for each token. Nvidia says Arcee trained Trinity Large on 2,048 Blackwell Ultra GPUs and prices hosted output at about $0.90 per million tokens. That makes paid hosted inference a current commercial route. Interconnects says Arcee previously earned revenue by post-training models for customer domains, then scaled down that work to focus on Trinity; executives described returning to monetization as the next step.
Vista Equity Partners, Cambium Capital and Emergence Capital led the round, according to Fortune. Microsoft’s M12, AI10 Ventures, Hitachi, IAG, P7 and Wipro also participated. Fortune said the proceeds would support additional open-weight models and products, expansion of Arcee’s Department of Energy partnership and work with Vista portfolio companies. Because the company did not state the amount, the $150 million figure remains reported rather than company-confirmed.
Arcee began by post-training other developers’ models, then moved upstream to pretraining its own. Fortune reported that the company spent about $20 million over several months to train four open-weight models, including Trinity Large. Nvidia says an earlier Trinity Large preview served 3.37 trillion tokens through OpenRouter in its first two months. That figure measures usage through a distribution platform; the sources do not disclose corresponding revenue or current enterprise contract volume.
The financing gives Arcee more room to repeat an expensive training cycle and build paid products around its models. Its permissive license also limits conventional lock-in because users can download the weights and operate them elsewhere. The business test is whether that reach produces enough paid API demand and renewed enterprise work to finance successive models as other open-weight labs keep releasing alternatives.