Boltz

Boltz is a public benefit corporation in Cambridge, Massachusetts, founded by the MIT researchers behind the open-source Boltz-1, Boltz-2 and BoltzGen biomolecular models. It is privately held, with no ticker. Unlike the drugmakers in the catalog's pharma section, it is an AI vendor: it sells access to its models through a hosted platform called Boltz Lab and a developer interface called the Boltz API, and it keeps publishing open models that anyone can download and run.

Its defining pattern is open-core. The research models that made its name are released under the MIT license, which covers both the code and the trained weights and permits commercial use: Boltz-1 in November 2024, Boltz-2 in June 2025, built with Recursion and able to estimate how tightly a molecule binds its target, and BoltzGen in October 2025, which designs new protein binders. The company says the models have been downloaded more than five million times. On top of them it sells proprietary pipelines. In June 2026 it launched BoltzMol-1 for finding small-molecule starting points and BoltzProt-1 for protein design, reached only through the API and Boltz Lab, with predictions priced from about two and a half cents each.

Boltz launched publicly in January 2026 with a 28 million dollar seed round from Zetta Venture Partners, Amplify Partners and Andreessen Horowitz. The same day it announced a strategic collaboration with Pfizer, under which Boltz refines Boltz-2 and BoltzGen on Pfizer's historical data to build exclusive models for structure prediction, binding affinity and biologics design. Similar deployments followed with Takeda in June 2026 and GSK in July 2026, and in September 2026 the flavor and fragrance company dsm-firmenich signed on to study scent and taste receptors. No financial terms were disclosed for any of them. In June 2026 Boltz also made its API available inside Anthropic's Claude products, so an AI agent can call the models directly.

Its most concrete published result is still the company's own. In September 2026 Boltz reported that BoltzMol-1 found three compounds with measurable activity against Werner helicase, a cancer target, after scoring about 172,000 purchasable compounds on a computer and testing only 27 in the lab, against three earlier screening campaigns that covered 350,000 compounds. The work was run by one Boltz chemist with a contract research lab, is posted as a preprint rather than peer reviewed, and the hits are weak starting points in the micromolar range, far from a drug.

The honest framing: Boltz's value rests on models that scientists already use widely and for free, and its business is a bet that drug and chemical companies will pay for accuracy, managed computing and custom models trained on their private data. The open models are a genuine commercial-use release, but the newest pipelines are closed, and hosted outputs may not be used to train competing models, so the open-source reputation and the paid product are no longer the same thing. Pfizer, Takeda and GSK signing on signals interest, not a measured discovery outcome. It sits between Chai Discovery, which licenses its newest antibody-design models to pharma rather than releasing them, and Isomorphic Labs, which keeps its drug design engine private: Boltz is open at the research layer and paid at the product layer.