Reflection AI Unveils Beam as Lower-Compute Rival to Chinese Open Models
Reflection AI has launched Beam, its first open-weight AI model, claiming it matches leading Chinese models in reasoning at lower costs and outperforms Western competitors. The text-only model aims to serve enterprises and sovereign nations, leveraging significant funding and compute deals to build 'AI factories'.
Reflection AI, a Brooklyn-based startup founded in 2024 by former Google DeepMind researchers, has unveiled Beam, its first frontier open-weight AI model, positioning it as a lower-cost competitor to leading Chinese open models such as DeepSeek, Qwen, and Z.ai. Beam is a 501-billion-parameter sparse mixture-of-experts model with 23 billion active parameters, designed for coding, reasoning and agentic workloads.
Reflection says the model can deliver competitive performance with 3–4 times less inference compute than some rivals, although those performance and efficiency claims have not yet been independently verified.
Beam was pretrained on 23.8 trillion tokens and subjected to a large-scale reinforcement-learning programme that Reflection says generated more than 100 million rollouts on 10,500 Nvidia GB300 GPUs over four weeks. The company says Beam is competitive with Z.ai’s GLM-5.2 and approaching Qwen 3.8-Max on coding and agentic tasks, while its own benchmark results show it ahead of Inkling, the open model developed by Mira Murati’s Thinking Machines Lab, on four coding tests where both models reported scores.
However, Reflection’s own evaluation also shows that models such as Kimi K3 and DeepSeek V4.1 Flash remain ahead on several raw capability benchmarks, underscoring that Beam’s principal pitch is inference efficiency rather than outright benchmark dominance.
The launch forms part of Reflection AI’s broader strategy to sell AI infrastructure to enterprises, developers and sovereign institutions, including its proposed “AI factory” model for organisations that want to build customised AI systems using their own data. The company has raised roughly $4.7 billion, according to PitchBook data cited by TechCrunch, and has secured more than $7 billion in computing agreements with SpaceX and Nebius for access to Nvidia GB300 chips through 2029.
Beam is currently undergoing final red-teaming and evaluation, with its weights, technical report, model card, and developer materials expected later this month, meaning developers cannot yet download the full model; Reflection plans to distribute it through hyperscalers, neoclouds, and open-source libraries once released.