Reflection Launches 501B-Parameter Beam to Rival Chinese Open Models

Reflection AI on Monday unveiled Beam, a 501-billion-parameter open-weight model it says matches leading Chinese open models on advanced reasoning benchmarks while running on a fraction of their compute. The debut positions the two-year-old startup against Z.ai, DeepSeek, and Qwen — and against U.S. frontier labs Anthropic and OpenAI.

How Beam Compares to the Field

Beam is a text-only mixture-of-experts model with 23 billion active parameters. Reflection pre-trained it on 23.8 trillion tokens and gave it a 1 million-token context window. By comparison, Z.ai's GLM 5.2 carries roughly 744 billion total parameters and 40 billion active parameters.

Reflection said the model is tuned for reasoning, coding, and agentic tasks. In a Monday blog post, the company described it as delivering performance "at a fraction of the token cost and inference time compute" of rivals.

What Performance Claims Has Reflection Made?

Reflection's benchmarks have not been independently verified. The company reports Beam scores on par with GLM 5.2 on advanced reasoning tests and outperforms current leading Western open models while consuming "3-4x less inference compute." Reflection calls Beam a "workhorse model" built for enterprises, the public sector, and developers.

Beam also outscores Inkling — the multimodal open model Mira Murati's Thinking Machines Lab released in July — on four coding benchmarks where both companies report results. Beam is text-only.

Who Is Backing Reflection?

The startup was founded in 2024 by two former Google DeepMind researchers. It has raised roughly $4.7 billion from Nvidia, Sequoia Capital, and Lightspeed Venture Partners, according to PitchBook. Its last funding round priced the company at a $25 billion pre-money valuation.

How Is Reflection Securing Compute?

This summer, Reflection signed deals worth more than $7 billion with SpaceX and Nebius to lock in access to Nvidia's GB300 chips through 2029. Compute capacity has become a decisive resource for startups racing to lure customers away from Anthropic and OpenAI's closed systems and China's cheaper open-weight releases.

What Is the AI Factory Strategy?

Reflection is pitching Beam and future models to enterprises and sovereign governments. The product concept, branded as "AI factories," would let institutions train Reflection's models on their own proprietary data to build customized local AI systems. Nvidia CEO Jensen Huang, whose company is a Reflection backer, has championed the same vision and stands to benefit if his GPUs power those systems.

Reflection has begun testing the sovereign AI factory approach with South Korea's Shinsegae Group, the company said. Axios previously reported that hedge funds and trading firms are among the parties interested in such setups.

When Will Beam Weights Be Available?

Reflection plans to release Beam's weights and full technical details this month. Distribution will run through hyperscalers and neoclouds, with integrations across open-source libraries at launch.

What Comes Next for Reflection?

The launch lands at a moment when Western open-source efforts are scrambling to keep pace with Chinese releases. Reflection's pitch — competitive reasoning at lower inference cost, paired with a sovereign-aligned deployment story — could make it a focal point for enterprise buyers evaluating Chinese alternatives. Whether independent benchmarks validate the claims will determine how quickly the $25 billion-valued startup converts that positioning into paying contracts.