Reflection AI has unveiled Beam, its first open-weight AI model, in a launch that takes direct aim at China’s open-model leaders. The company announced the model on October 5.

Beam is a sparse mixture-of-experts system with 501 billion total parameters, of which only 23 billion are active for any given token. Reflection says it was pretrained on 23.8 trillion tokens and built for coding, reasoning and agentic workloads.

The company published self-reported scores of 80.9 on SWE-Bench Verified, 80.1 on Terminal-Bench 2.1 and 97.8 on AIME 2026. These figures come from Reflection and had not been independently verified at launch. The firm claims Beam needs three to four times less inference compute than Z.ai’s GLM-5.2.

The full weights, technical report and model card will ship under an Apache 2.0 license later in October. Early access is currently open through a waitlist. The Nvidia-backed startup was founded in March 2024 by DeepMind alumni Misha Laskin and Ioannis Antonoglou. It has raised funds at a reported valuation near $25 billion and signed over $7 billion in compute deals with SpaceX and Nebius for access to Nvidia’s GB300 chips through 2029.