Reflection debuts Beam, a 501B-parameter open-weight MoE model focused on coding, reasoning, and agentic workloads

Reflection introduced Beam, its first open-weight model, a sparse Mixture-of-Experts system with 501B total parameters and 23B active parameters. The company says Beam was pretrained on 23.8T tokens and further trained with large-scale reinforcement learning using 10.5K NVIDIA GB300 GPUs, with weights and technical materials planned for release later this month.

Beam highlights the growing push to make frontier-style capabilities more accessible through efficient open models rather than only larger closed systems. Its emphasis on inference efficiency and agentic coding suggests open-weight competition is shifting from raw benchmark scores toward practical deployment costs and developer workflow performance.