AI
Reflection unveils Beam, a 501B open-weight model pitched against Chinese labs
Reflection AI introduced Beam, its first frontier open-weight model, on October 5. Beam is a sparse mixture-of-experts system with 501 billion total parameters that activate about 23 billion per token, pretrained on 23.8 trillion tokens with a one-million-token context window. The Nvidia-backed startup says it matches leading Chinese open models such as Z.ai's GLM-5.2 on advanced reasoning benchmarks at roughly a quarter of the inference cost. Weights and a technical report are promised later this month.