Reflection AI Unveils Beam, Open-Weight Model Challenging Chinese Rivals at Lower Compute Cost

Reflection AI, a startup based in Brooklyn, has introduced Beam, its inaugural open-weight AI model, asserting that it matches the performance of top Chinese models on advanced reasoning tests while requiring less computational power. Established in 2024 by ex-Google DeepMind researchers, Beam features a mixture-of-experts architecture with 501 billion parameters, of which 23 billion are active, and has been pretrained on 23.8 trillion tokens. It excels in reasoning and coding, surpassing Western models while consuming 3-4 times less inference compute. Reflection has secured $4.7 billion from investors including Nvidia and Sequoia Capital, boasting a pre-money valuation of $25 billion. This summer, it finalized over $7 billion in agreements with SpaceX and Nebius. The weights and technical specifications for Beam will be disclosed this month.