Reflection claims Beam matches GLM-5.2 with 3-4x less computing power

Reflection AI says its Beam model equals Z.ai's GLM-5.2 on advanced reasoning tests while needing 3-4x less computing power to run answers. Beam is a text-only mixture-of-experts design, where only part of the model works at once, with 501 billion total parameters and 23 billion active, versus about 744 billion and 40 billion for GLM-5.2.