Nvidia-Backed Reflection AI Launches Beam, a 501B-Parameter Open Model to Challenge Chinese AI

Reflection AI, a New York City startup founded in March 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou, has unveiled its first open-weight model, called Beam. Backed by Nvidia and positioning itself as America's answer to open-source Chinese AI, the company says Beam is designed as a "workhorse" particularly suited to coding and AI-agent tasks.

Beam is a 501-billion-parameter model that Reflection claims matches or approaches leading Chinese open models. The company benchmarked it as comparable to Z.ai's GLM-5.2 and approaching Alibaba's Qwen 3.8-Max in performance. Crucially, Reflection says Beam requires three to four times less computing power to reason through a problem than comparable Western open models, making it significantly cheaper to run for businesses and developers.

The company attributes Beam's efficiency to high-compute reinforcement learning, which it says allows competitive performance on coding and agentic tasks at a fraction of the token cost and inference time compute of rival models. Institutions such as hedge funds and trading firms are reported to already be interested in using the model to build proprietary AI systems on their own data. The full model release is expected later in October 2026.