Reflection AI Launches Beam, a New U.S. Open-Weight Model for Businesses and Agencies

Beam is a text-only mixture-of-experts model with 501 billion total parameters, 23 billion active parameters, pre-trained on 23.8 trillion tokens, and a context window of 1 million tokens.
Reflection has partnered with the U.S. Department of Energy and the Pentagon, and has agreements to develop AI models for U.S. allies including South Korea.
Reflection has raised more than $4 billion and reached a reported $25 billion valuation before releasing its first model.
Open-weight models’ share of tokens processed through Vercel’s AI Gateway rose to 56% in August, from 7% the previous December, underscoring their growing use.
Open models’ easier access and customizability also raise security concerns: U.S. and U.K. government evaluators found recent Chinese models could help hackers exploit software vulnerabilities.
Reflection AI unveiled Beam, its first open-weight AI model, positioning the U.S. startup as a competitor to leading Chinese models. The Hill reports that the text-only model contains 501 billion total parameters with 23 billion active parameters, trained on 23.8 trillion tokens over four weeks. TechCrunch notes Reflection claims Beam matches the performance of Chinese open models while using three to four times less computing power to run.
The Brooklyn-based startup has already attracted major backing. Yahoo Finance reports Reflection has raised over $4 billion and reached a $25 billion valuation before releasing its first model. The company has partnered with the U.S. Department of Energy and the Pentagon, and signed agreements to develop AI models for U.S. allies including South Korea.
Open-weight models are becoming mainstream. Digg reports that open-weight models' share of tokens processed through Vercel's AI Gateway jumped to 56% in August, up from just 7% the previous December. This dramatic shift shows developers and businesses are increasingly adopting customizable models they can run on their own infrastructure.
Unlike cloud-based AI services, Beam can run entirely on users' own servers. This means companies keep sensitive data in-house and avoid recurring API fees. Yahoo Finance notes that Reflection is marketing this flexibility to businesses, developers, and public agencies seeking alternatives to proprietary Chinese models controlled by distant companies.
The easier availability of open models creates new dangers. U.S. and U.K. government evaluators found that recent Chinese open-weight models could help hackers discover and exploit software vulnerabilities. Yahoo Finance reports that Reflection's model faces the same scrutiny, as regulators balance the benefits of competition against national security concerns about unrestricted AI access.
Beam is just the beginning. TechCrunch reports that Reflection is already training a more powerful successor model. The company's rapid development pace reflects its mission: create an American alternative to Chinese open models that dominates the market for public, customizable AI tools.
Publishers
12
Articles
9
Reach
21