Recently, after chatting with Percy Liang—Stanford professor and leader of the MARIN project (yet another fully open model lab)—it struck me that, in time, a coalition of companies will likely come together to fund a core set of foundational open models that become widely adopted across the industry. It’s uncertain when this will materialize, and while Nemotron (Coalition) represents Nvidia’s effort to fund and kickstart the approach inside one rich company, a consortium remains the only sustainable long-term route to well-resourced, near-frontier open models. In recent months, open-model labs have seen considerable upheaval, including high-profile departures at Qwen and Ai2 (my comment). This shouldn’t come as much of a surprise to anyone following the space — it’s already happened with Meta deprioritizing Llama, and it will only become more common as the expense of staying competitive at the AI frontier continues to climb. Other major labs offering models today include Chinese startups like Moonshot AI, MiniMax, and Z.ai — all of which appear vulnerable in their ability to sustain the rising costs of training and R&D. Releasing top-tier models openly now conflicts with prioritizing resources toward AI products that can generate real revenue and profits today. We’ll see new business models built around openly releasing some—or even many—AI models. These will mostly be smaller models that support a wide variety of niche use cases, rather than cutting-edge models at the leading edge. This class of companies that’ll release many, strong fine-tunable models will include the likes of Arcee AI, Thinking Machines, OpenAI, Google with Gemma, and more in that class. The financial cost and competitive edge of keeping top models closed-source are simply too great in a business landscape filled with revenue-generating opportunities.
Interconnects AI