Reflection AI launches Beam open-weight model for coding

Reflection AI, a startup backed by Nvidia, has launched Beam, an open-weight artificial intelligence model designed for coding and agentic tasks. The model aims to provide a Western alternative to low-cost Chinese models, such as Z. ai’s GLM-5.2 and Alibaba’s Qwen3.8-Max.

Beam utilizes a specialized architecture featuring 501 billion total parameters, but only activates 23 billion parameters for any individual task. This approach is intended to combine the high-level capabilities of much larger models with the lower operating costs and faster response times associated with smaller systems. By allowing developers to download and run the model parameters on their own hardware, Reflection AI offers an alternative to traditional API-based usage, shifting the focus toward custom integration and MLOps.