Arcee AI raises $150M at $1B valuation after training four models for $20M

Most AI startups burn through $20 million before their office espresso machine is fully calibrated. Arcee AI spent that amount training an entire family of large language models, then walked into a fundraising round that valued the company at over $1 billion. The San Francisco-based company, founded in June 2023 by Mark McQuade, Jacob Solawetz, and Brian Benedict, has raised $150 million to fuel what comes next: an open-weight model exceeding one trillion parameters. The round drew backing from the venture arms of Microsoft (M12), Hitachi Ventures, Samsung Next, Wipro, and Prosperity7 Ventures. Four models, $20 million, 30 people Arcee’s headline trick is its Trinity model family, a set of open-weight models spanning a wide range of sizes. The Nano variant runs 6 billion parameters. Mini comes in at 26 billion. The flagship Large/Thinking model clocks 400 billion parameters total, with roughly 13 billion active at any given time thanks to a mixture-of-experts architecture. Mixture-of-experts, or MoE, is essentially a routing system: instead of firing every neuron on every query, the model activates only the subset of “expert” sub-networks most relevant to the task. The result is a ...