Nvidia-backed start-up Reflection unveils open AI model to rival China offerings
The model is designed to perform for business customers and developers on coding and AI agent tasks.
San Francisco - Reflection AI, a startup founded by two former Google DeepMind researchers, has unveiled an open-weight AI model named Beam. The company claims Beam can compete with leading models in the United States and China, particularly for business and coding tasks, while being more cost-effective. Beam's performance matches that of GLM-5.2, an open model developed by China's Z.AI, on advanced reasoning benchmarks. The startup plans to release further technical details about Beam later this month.
China has been making significant strides in AI, with developers such as DeepSeek and Moonshot AI offering affordable and customizable open models. In contrast, major US tech companies like OpenAI and Anthropic have primarily focused on closed or proprietary models to bolster their enterprise businesses. Nvidia, along with other investors, backed Reflection AI as it emerged from stealth mode in 2025.
The startup has been in discussions to raise $2.5 billion at a valuation of $25 billion, according to a previous report by the Wall Street Journal. Reflection has also secured billion-dollar agreements with SpaceX and Nebius Group to access computing resources for model development. The model's release is expected to bolster the United States' efforts to export AI technology abroad, including plans for sovereign AI projects in Korea.
The company's success may pave the way for more US tech exports, aiming to counter the proliferation of Chinese alternatives.
Written by urgent.news from Straits Times Business's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
