October 6, byNvidiaInvesting in start-ups Reflect AI launched a major model of home-based open rights on Monday, local time, Beam. The start-up company wants to compete with other Chinese AI models like DeepSeek and Kimi in the field of code writing and intelligent mission。
Reflect AI indicates that Beam is able to perform a GLM – 5.2 for China ' s AI Initial Enterprise Intelligence Speculation AI (Z.ai); in terms of code and intelligent body tasks, the model 's performance is coming closer and closer to the question 3.8 - Max (Qwen 3.8 - Max)。
The start-up company claims that the total parameter volume of Beam is 50.1 billion; that's the variable that determines how the AI system handles information. But the model is a thin-activated structure, and each processing task only activates 23 billion parameters. Only one part of the network is activated each time to achieve faster reasoning while reducing operational costs。
BY CONTRAST, THE GLM-5.2 TOTAL PARAMETER OF AI IS APPROXIMATELY 74.4 BILLION AND THE ACTIVATED PARAMETER IS 40 BILLION。
As the model was released, United States technology firms were trying to resist competitive pressures from China’s open weight model. These models are cheaper and more customized, and the coding capacity is almost capable of catching up with OpenAI and Anthropic head models。
According to the IT House, Reflecting AI was created in 2024 by former DeepMind researcher Misha Laskin and Ioannis Antonoglou. The company has mainly developed software development automation tools and software development automation is a very fast-growing type of AI landing scene。
Earlier this year, Reflecting AI has entered into a cooperative agreement with SpaceX to obtain additional calculator resources at the SpaceX 2 data centre。