EvolutionaryScale

What EvolutionaryScale is

EvolutionaryScale is an AI research company that introduced itself to the public on 25 June 2024, the same day it published a model called ESM3 ^[1]. Its own description of itself is short and specific: a public benefit company whose mission is to develop AI that understands biology for the benefit of human health and society ^[1]. When Biohub later announced that the team was joining it, it used similar language, calling EvolutionaryScale a frontier AI research lab and public benefit company that had created large-scale AI systems for the life sciences ^[2].

A company and a model at the same time

The company's first public act was a model release. ESM3, it said, was an order of magnitude larger than ESM2, the team's previous model ^[1]. Announcing a company together with a flagship system tells readers what kind of organisation it intends to be: the research output is the thing the organisation is built around, and the company exists to keep producing it. The comparison with ESM2 also places the new work in a line of succession — the same team, a bigger model, the next step rather than a fresh start ^[1].

The responsible development framework

Alongside the model, EvolutionaryScale published a responsible development framework ^[1]. The framework was drawn up with reference to the scientists' own research experience, and it commits the company to transparency and accountability ^[1]. Publishing such a document at launch, rather than after a controversy, is a way of stating in advance how the company intends to behave as its systems grow more capable, and of giving outsiders something concrete to hold it to.

How the model was made available

At launch, ESM3 was offered through an API in closed beta, and the company announced availability through AWS SageMaker, NVIDIA BioNeMo and an NVIDIA NIM microservice ^[3]. It also released a small open model, ESM3-open, with weights and source code on GitHub ^[3]. The two routes served different users: a hosted API and cloud integrations let people try the largest model without managing it themselves, while downloadable weights and source code let others inspect and build on a smaller version. For a company arguing that its technology should be scrutinised, shipping an open model alongside a closed one is itself part of the argument.

Joining Biohub

On 6 November 2025, Biohub announced that the EvolutionaryScale team would join it ^[2]. Co-founder and chief scientist Alex Rives became Biohub's head of science ^[2]. Rives is a core institute member at the Broad Institute and an assistant professor in MIT's Department of Electrical Engineering and Computer Science ^[2]. The company's website now states that it is part of Biohub and that its models, APIs and services are available at biohub.ai ^[3].

Where it sits

The company's short history follows a recognisable arc for frontier AI labs: an independent launch built around one large model and a public statement of principles, then absorption into a larger scientific institution that can host the work. What carried over was the model line and the services around it, now served from Biohub's site ^[3]; what changed was the institutional setting, with the team's chief scientist taking a senior science role at the organisation it joined ^[2]. For readers following how biological AI is organised, EvolutionaryScale is a case of a small, mission-framed company being folded into established research infrastructure rather than growing into a large independent firm.