Boltz Open-Source AI Models Win Pfizer Deal and $28M Seed

What happened

Boltz PBC, a young AI lab building open-source models of how biological molecules interact, has had a busy start to 2026. The company closed a $28 million seed round and announced a collaboration with Pfizer to apply its models to drug design 123. The round was co-led by a16z, Amplify Partners, and Zetta Venture Partners and valued the company at $125 million post-money 1.

The Pfizer agreement, announced in January 2026, centers on two Boltz systems. Boltz-2 predicts how proteins and small molecules bind. BoltzGen is aimed at generative design. Pfizer plans to use them across both small-molecule and biologics programs 3.

Who Boltz is

Boltz was founded in late 2024 by MIT researchers Gabriele Corso, Jeremy Wohlwend, and Saro Passaro 1. It is organized as a public-benefit corporation, and its stated mission is to put capable biological AI tools in the hands of scientists everywhere 1. One profile lists the company's headquarters as London, while trade coverage emphasizes its MIT roots 12.

The company's reputation rests on its models. Boltz-1 is an open co-folding model that predicts the structures of interacting biomolecules. It reportedly comes close to the accuracy of proprietary systems such as AlphaFold3 while remaining freely available to researchers 1. Boltz-2 adds binding-affinity prediction and is described as state of the art in both speed and precision 1. That kind of estimate sits close to the commercial core of small-molecule discovery.

Why openness is the story

The accounts largely agree on the facts but frame the company's origins a little differently. Endpoints ties the founding directly to Google DeepMind's decision to release AlphaFold 3 in a closed form, casting Boltz as a response to that moment 2. The Grokipedia profile takes a wider view and positions Boltz against a general drift toward proprietary models in computational biology 1. IntuitionLabs goes further still, presenting open biomolecular foundation models as part of a broader paradigm shift in how pharma approaches discovery 3.

These framings fit together. AlphaFold 3's restricted release was the clearest example of a trend in which the most capable structure-prediction tools were pulling away from academic and open access. Boltz's argument is that an open alternative can stay close enough to the frontier to matter. The Pfizer deal is the first major signal that a large drugmaker agrees.

Why it matters

For pharma, open models offer things closed APIs struggle to match. Companies can inspect the model, fine-tune it on proprietary data, and run it inside their own infrastructure without sending sensitive chemistry to an outside service. That logic matters more as models move from predicting structures to scoring binding and generating candidates, which are steps closer to decisions about real molecules.

For Boltz, the combination of venture backing and a marquee partner tests a business model that the field has not fully proven. The company gives away its core models, so its revenue presumably depends on what it builds around them: tooling, partnerships, and specialized deployment. A $125 million valuation at seed suggests investors believe that layer can be large 1. The public-benefit structure signals that openness is meant to be a lasting commitment rather than a launch tactic 1.

Is Pfizer a one-off?

The obvious question is whether Pfizer is an exception or the first of several pharma partners. The available reporting documents only the Pfizer collaboration 23. Claims about additional big-pharma partners should be treated with caution until those companies or Boltz confirm them.

Still, the strategic case for repeat deals is strong. An open model is not exclusive to any single partner, so one company's adoption does not lock out its competitors. Each new integration also makes the model more valuable to the next. If Boltz-2 and BoltzGen perform in Pfizer's pipelines as advertised, other drugmakers that have hesitated to depend on closed systems have a ready alternative.

The takeaway

Boltz has turned an ideological position, that biomolecular AI should be open, into a funded company with a top-tier pharma customer within roughly a year of its founding 123. The harder test comes next. The company must show that open models can keep pace with well-resourced proprietary labs, and that openness can sustain a durable business rather than just win early goodwill. The Pfizer deal does not settle either question, but it makes Boltz the clearest test case for the open approach in AI drug discovery.