Reflection AI boosts US ambition to compete with Chinese ‘open’ models

Reflection AI, a two‑year‑old startup based in Brooklyn, unveiled its first open‑weight model, Beam, on Monday, positioning it as a direct Western competitor to leading Chinese open models such as DeepSeek, Qwen and Z.ai. Beam is a 501‑billion‑parameter mixture‑of‑experts system with 23 billion active parameters, trained on 23.8 trillion tokens and featuring a one‑million‑token context window. The company claims the model matches the performance of China’s GLM‑5.2 on advanced reasoning benchmarks while using three to four times less inference compute, delivering “a fraction of the token cost and inference time compute” of rivals. Beam is marketed as a workhorse for enterprises, the public sector and developers, and Reflection plans to release its weights and full technical details later this month through hyperscalers, neoclouds and open‑source integrations.

The launch underscores Reflection’s broader strategy to secure a foothold in the escalating AI race by leveraging massive compute resources and deep‑pocketed financing. Backed by Nvidia, Sequoia Capital and Lightspeed Venture Partners, the startup has raised roughly $4.7 billion and was valued at a $25 billion pre‑money valuation in its latest round. This summer it signed more than $7 billion in agreements with SpaceX and Nebius to access Nvidia’s GB300 chips through 2029, ensuring the compute power needed to train frontier models. Reflection’s pitch centers on “AI factories,” allowing enterprises and sovereign nations to train customized, locally hosted AI systems on proprietary data—a concept championed by Nvidia CEO Jensen Huang that could attract hedge funds, trading firms and other data‑intensive organizations.

While Reflection’s performance claims have not yet been independently verified, the company’s internal benchmarks show Beam outscoring Inkling, Mira Murati’s open model, on four coding tests, despite Inkling’s multimodal capabilities. Reflection has already begun testing sovereign AI‑factory partnerships, notably with South Korea’s Shinsegae Group, and is targeting both U.S. and international customers seeking alternatives to closed‑source offerings from Anthropic and OpenAI. The forthcoming public release of Beam’s weights could intensify competition in the open‑model ecosystem, potentially reshaping how enterprises and governments procure and deploy large‑scale AI, while also bolstering Nvidia’s GPU market through increased demand for its hardware.

⚡ Effects Interpreter