Brooklyn-based startup Reflection AI has launched Beam, a 501-billion-parameter open-weight model designed to compete with Chinese and Western AI models at lower compute costs.

Reflection AI has officially debuted Beam, its first frontier, open-weight AI model. The two-year-old, Brooklyn-based startup claims the model matches the performance of leading Chinese open models on advanced reasoning benchmarks while utilizing significantly less inference compute. The company intends to release the model's weights and technical details this month through various hyperscalers and open-source libraries.

Technical Specifications and Performance

Beam is a text-only mixture-of-experts model featuring 501 billion total parameters, with 23 billion active parameters. It was pre-trained on 23.8 trillion tokens and supports a 1 million token context window. According to internal benchmarks, Reflection claims Beam performs on par with Z.ai’s GLM-5.2 model while using three to four times less inference compute. The company reports that Beam also outperforms existing Western open models and the Inkling model from Thinking Machines Lab on specific coding tests.

Business Strategy and AI Factories

Founded in 2024 by former Google DeepMind researchers, Reflection has raised approximately $4.7 billion, reaching a $25 billion pre-money valuation. The startup is targeting enterprises, the public sector, and sovereign nations with a strategy centered on "AI factories." This concept allows institutions to train Reflection’s models on their own proprietary data to create customized, local AI systems. To support its infrastructure needs, the company secured deals worth over $7 billion with SpaceX and Nebius for access to Nvidia GB300 chips through 2029.