Nvidia-backed Reflection unveils first AI model to take on Chinese open models

Nvidia-backed startup Reflection AI unveiled its first open-weight AI model, called Beam, on Monday. The company aims to challenge Chinese open models like DeepSeek and Kimi in the domains of coding and agentic tasks. Beam is positioned to be competitive with Z.ai's GLM-5.2 and is rapidly approaching Qwen3.8-Max in performance for coding and agentic tasks.

Beam boasts a substantial 501 billion total parameters, or variables that influence an AI system's information processing. However, the model only utilizes 23 billion parameters for each individual task, optimizing performance and cost-efficiency by selectively activating a portion of its network. In contrast, Z.ai's GLM-5.2 features around 744 billion total parameters with 40 billion active parameters.

This launch follows a trend among US tech firms striving to counter the rise of Chinese open-weight models. These models offer lower costs, greater customizability, and code generation abilities that rival leading models from OpenAI and Anthropic. Founded in 2024 by former DeepMind researchers Misha Laskin and Ioannis Antonoglou, Reflection AI develops tools designed to automate software development, a rapidly expanding application for AI.

In a notable partnership, Reflection AI secured a deal with SpaceX this year, securing additional computing resources at the Elon Musk-led company's Colossus 2 data center.