Reflection AI has officially unveiled Beam, its first frontier, open-weight AI model. The Brooklyn-based startup, founded in 2024 by two former Google DeepMind researchers, claims that Beam matches the performance of leading Chinese open models on advanced reasoning benchmarks at dramatically lower costs. This announcement could intensify the competition to develop a Western alternative to models like DeepSeek, Qwen, and Z.ai.

Key Features of Beam

In a lengthy blog post released on Monday, Reflection described Beam as a text-only mixture-of-experts model trained on high-compute reinforcement learning. It is designed to excel at reasoning, coding, and agentic tasks while operating at “a fraction of the token cost and inference time compute” compared to its rivals.

  • Model Size: Beam is a 501-billion-parameter model with 23 billion active parameters.
  • Training Data: It was pre-trained on 23.8 trillion tokens and features a 1 million token context window.
  • Performance: Reflection claims that Beam scores on par with Z.ai’s GLM-5.2, which has roughly 744 billion total parameters with 40 billion active.

Competitive Landscape

While Reflection's performance claims have not been independently verified, the company asserts that Beam outperforms today’s leading Western open models while using “3-4x less inference compute.” Reflection is positioning itself against closed labs like Anthropic and OpenAI, as well as popular open models from Chinese developers and Western players like Mistral, Meta, and Cohere. Its most direct U.S. rival may be Inkling, the open model from Mira Murati’s Thinking Machines Lab, released in July. Reflection's benchmarks indicate that Beam outperforms Inkling on four coding tests where both report results, although Inkling is a multimodal model while Beam is text-only.