The AI race just opened another front. Reflection AI has released Beam, a 501 billion parameter open weight model designed to challenge China’s increasingly strong position in open AI. Beam is being positioned for coding and agentic workloads, with Reflection claiming competitive performance at substantially lower inference compute. But the model itself is only part of the signal. Nvidia has invested roughly $800 million in Reflection. The company is already working with the Pentagon and Department of Energy. And because Beam uses an Apache 2.0 license, organizations can modify it and deploy it on infrastructure they control. That matters. The United States has built some of the world’s strongest proprietary AI systems, while Chinese labs have become increasingly formidable in downloadable models. Governments and companies increasingly want more than API access. They want control over the model, the data and the infrastructure underneath it. Beam’s performance still needs indepe...