Overview
- Reflection publicly unveiled Beam on Monday, announcing it as the startup’s first frontier open-weight model and saying full weights and technical docs will be released later this month.
- Beam is a sparse mixture-of-experts model with 501 billion total parameters that activates about 23 billion parameters per token, was pretrained on 23.8 trillion tokens, and supports a 1 million token context window.
- The company claims Beam scores on par with Z.ai’s GLM-5.2 on advanced reasoning and coding tests while using roughly 3–4 times less inference compute, but those benchmark results are self-reported and await independent verification.
- Reflection has secured large GPU capacity through multibillion-dollar compute deals with SpaceX’s Colossus and Nebius and has Nvidia financial and technical backing to push its commercial ‘AI factory’ pitch to enterprises and governments.
- Whether Beam reshapes the market will depend on the upcoming public weight release, the Apache 2.0 license terms, third-party benchmarks, enterprise signups, and scrutiny of security and regulatory risks.