vLLM

Reflection releases Beam: 501B open MoE model with Apache 2.0

On October 16, 2026, Reflection released Beam, a 501B-parameter mixture-of-experts model under Apache 2.0, with 23B active parameters per token and a 1M context window.