Nvidia-backed start-up Reflection AI has officially launched its first open-weight frontier model, Beam, claiming it matches the performance of leading open-weight competitors from China while offering lower running costs.

The text-only, mixture-of-experts model is aimed at reasoning, coding and agentic tasks.

It includes a total of 501 billion parameters, but only 23 billion are active at any given time, making for relatively compute-efficient operations.

Reduced operating costs

By comparison, the GLM-5.2 model from Beijing-based Z.ai, also known as Zhipu AI, has roughly 744 billion parameters, with 40 billion active.

Reflection’s claims haven’t yet been independently verified, but it said benchmark tests showed Beam was on par with GLM-5.2 and outperformed leading open models from Western countries with three to four times less inference compute.

The most advanced open-weight models are developed by Chinese firms including Z.ai, Moonshot and Alibaba, but France’s Mistral, Facebook parent Meta, and Canada’s Cohere are among the Western companies with open-weight offerings.

Open-weight models

Brooklyn-based Reflection was founded in 2024 by two former Google DeepMind researchers, and has raised roughly $4.7 billion (£3.5bn) from backers including Nvidia, Sequoia Capital and Lightspeed, valuing it at $25bn pre-money before the release of its first product.

Nvidia, the US administration and others have called for a stronger Western ecosystem of open-weight models to balance the proprietary offerings from OpenAI and Anthropic.

Such offerings allow companies to run models on their own hardware and to fine-tune them for their own needs.