Reflection launches Beam, an open-weight model built for lower-cost reasoning

Brooklyn-based startup Reflection has unveiled Beam, its first frontier open-weight model. Beam is a text-only mixture-of-experts model with 501 billion total parameters and 23 billion active parameters. It was pretrained on 23.8 trillion tokens and supports a 1 million-token context window.

According to Reflection, Beam performs on par with Z.ai's GLM-5.2 on advanced reasoning benchmarks and outperforms leading Western open models while using 3-4x less inference compute. The claims have not been independently verified. The company describes Beam as a "workhorse model" for enterprises, the public sector, and developers.

Founded in 2024 by two former Google DeepMind researchers, Reflection has raised roughly $4.7 billion from investors including Nvidia, Sequoia Capital, and Lightspeed Venture Partners. In summer 2026, the startup signed compute deals worth more than $7 billion combined with SpaceX and Nebius to secure access to Nvidia GB300 chips through 2029.

Reflection said it will release Beam's weights and full technical details in October 2026, with distribution through hyperscalers and neoclouds and integrations across open source libraries at launch. The company has also begun piloting a sovereign AI factory partnership with South Korea's Shinsegae Group.