Reflection unveils Beam, an open-weight AI model with lower compute costs

Reflection AI has unveiled Beam, its first open-weight model. The two-year-old, Brooklyn-based company says Beam matches leading Chinese open models on advanced reasoning benchmarks while requiring much less inference compute. Those performance claims have not been independently verified.

Beam is a text-only mixture-of-experts model with 501 billion total parameters and 23 billion active parameters. It was pre-trained on 23.8 trillion tokens and has a 1 million-token context window. Reflection says the model is designed for reasoning, coding, and agentic tasks, and uses 3–4 times less inference compute than rivals. For comparison, Z.ai’s GLM-5.2 has roughly 744 billion total parameters and 40 billion active parameters.

Reflection says its benchmarks put Beam on par with GLM-5.2 for advanced reasoning and ahead of leading Western open models. The company also reports that Beam scores higher than Inkling on four coding tests where both models have results; Inkling is multimodal, while Beam handles text only.

Reflection says it will release Beam’s weights and full technical details this month, with distribution through hyperscalers and neoclouds and integrations across open-source libraries at launch. The company is targeting enterprises and sovereign nations with a proposal for customized local AI systems trained on institutions’ own data.