Beam is an open‑weight model from Reflection. It uses a mixture‑of‑experts architecture that activates only 23 billion of its 501 billion parameters per token. Beam targets GLM 5.2 performance on coding and reasoning while cutting compute by 3‑4×. The release positions Reflection against Chinese rivals Deepseek and Qwen, emphasizing efficiency for developers seeking high‑performance models today.
Opening Kapyn…