Reflection AI Introduces Beam: A 501B Open-Weight MoE Model With 23B Active Parameters for Coding and Agentic Workloads

Reflection AI unveils Beam open-weight model to compete with Chinese AI

Nvidia-backed Reflection AI debuted Beam, a 501B parameter sparse MoE model with 23B active parameters, designed to compete with lower-cost Chinese models like DeepSeek in coding and agentic tasks.