⚡Reflection released Beam 1.0, an open 501B-parameter mixture-of-experts language model with 23 billion active parameters.

What it means

Reflection AI released Beam 1.0, its first open-weight large language model, on Hugging Face. It is a sparse mixture-of-experts (MoE) model with 501 billion total parameters but only 23 billion active per token, which makes it far cheaper to run than a dense 501B model. It is aimed at software engineering, agentic tool use, and reasoning, and the weights are released under Apache 2.0.

Accuracy & sourcesGrounded

The claim is solid: this is the first-party release itself, and the 501B total / 23B active MoE configuration matches Reflection's own blog and the Hugging Face model card. Reporting sits essentially at the primary source, so there is no meaningful gap between the claim and the evidence.

How new

Current, not resurfaced: the release and companion blog posts are dated October 16, 2026, the same day as this wave of coverage.

Past coverage