Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost
By Pradhyuman,
Reflection AI announced Beam, an open-weight, text-only mixture-of-experts model. The Brooklyn-based startup stated that Beam has 501 billion total parameters, 23 billion active parameters, and a 1 million token context window. Reflection trained the model on 23.8 trillion tokens. The company said Beam matches the reasoning performance of Z.ai's GLM-5.2 model and requires three to four times less inference compute.
Two former Google DeepMind researchers founded Reflection in 2024. According to PitchBook, the company raised about $4.7 billion from investors, including Nvidia, Sequoia Capital, and Lightspeed Venture Partners. Reflection signed compute deals worth over $7 billion with SpaceX and Nebius to access Nvidia GB300 chips through 2029. The startup pitches "AI factories" to let institutions train models on proprietary data. Reflection is testing this product with South Korea's Shinsegae Group and plans to release Beam's weights this month.
Maintained by Pradhyuman.
Filed under: Google, Nvidia, Models & Research, Funding & Business