Reflection AI

Reflection AI unveiled Beam on 5 October 2026, its first frontier open-weight model: a text-only mixture-of-experts model with 501B total and 23B active parameters, pre-trained on 23.8T tokens with a 1M-token context window. The company claims performance on par with Z.ai's GLM-5.2 on advanced reasoning benchmarks while using 3-4x less inference compute than leading Western open models; these claims are not independently verified. It is the first shipped product from the Nvidia-backed lab and the proof point behind its valuation and compute commitments.

Source: TechCrunch