Nvidia-backed Reflection AI Launches Beam Model to Challenge Chinese Rivals

Nvidia-backed startup Reflection AI has launched its first open-weight model, called Beam, aiming to challenge Chinese competitors like DeepSeek and Kimi in coding and agentic tasks. The model is designed to be faster and cheaper to run, with 501 billion total parameters but only 23 billion active for each task. Reflection claims Beam performs well against Z.ai’s GLM-5.2 and is nearing the capabilities of Qwen3.8-Max.

The launch comes as U.S. tech firms seek to counter the rise of cheaper, customizable open-weight models from China, which can write code almost as effectively as top models from OpenAI and Anthropic. Reflection AI was founded in 2024 by former DeepMind researchers Misha Laskin and Ioannis Antonoglou and has previously secured a deal with SpaceX for computing power at the Colossus 2 data center.

While not a direct earnings event for Nvidia, the AI model race is closely tied to the company’s stock performance. Nvidia’s CEO Jensen Huang has highlighted the success of open models, noting that nearly all run on Nvidia hardware. The company reported $96 billion in Q2 revenue, more than double the previous year, with data center revenue hitting $89 billion. Nvidia expects $108 billion in Q3 revenue and 70% growth next fiscal year.

Investors watching Nvidia stock will be keen to see how open models like Beam gain traction and how Chinese rivals respond. Cheaper models could drive more developers toward open tools, many of which rely on Nvidia chips. However, risks include a potential dip in gross margin to 71-72% in Q4 due to rising memory costs, and Nvidia’s outlook excludes China data center revenue.