A new open-weight A.I. model called Beam, introduced by Reflection AI, is emerging as a potential challenger to established players like Anthropic and Open AI. Beam is focused on coding and A.I. agents, applications popular with business users. Reflection claimed that Beam’s reasoning performance is on par with GLM-5.2, a powerful open-weight model by the Chinese start-up Z.ai, and outperforms many Western open-weight ones. Founded in 2024 by two former Google Deep Mind researchers, Misha Laskin and Ioannis Antonoglou, Reflection has raised billions of dollars from investors including Nvidia, Sequoia, Disruptive Technology Advisers and 1789 Capital. The company was most recently valued at $25 billion.
Open-weight models have become a big topic of debate in A.I. These tools, which users can modify, have risen in popularity as companies seek cheaper alternatives to closed-source models from places like Anthropic, Google and Open AI. Developers have increasingly adopted powerful open-weight models from Chinese labs, worrying leaders in Silicon Valley and in Washington.
This trend has stoked interest in similar tools developed by Western companies like Meta, Thinking Machines Lab, and Mistral. Reflection adds to the challenges confronting Anthropic, which is expected to go public next month amid growing questions about model safety and increasing competition.
The implications of a successful U.S. open-weight model are significant. If companies move away from more costly closed models, the economics of the business models of Anthropic and Open AI fundamentally weaken, potentially battering their valuations and affecting other big tech giants.
来源: New York Times Business · 由HeadlinesBriefing整理摘要