[DigitalToday reporter Chi-gyu Hwang] AI models marketed on value for money appear to be directly affecting sales of cutting-edge, high-priced models from Anthropic and OpenAI.

A recent Financial Times report citing an analysis by payments company Ramp of spending data from 70,000 companies showed Anthropic's top-tier, highest-priced model Fable 5 accounted for about 11 percent of spending on Anthropic tools, more than 2 months after its launch.

That differs from the previous pattern in which corporate users often chose the best-performing model as the default option.

The FT, citing analysts and Anthropic investors, said the model's high price and the view that existing models can cover most business needs contributed to the shift. It added that slower adoption of Fable 5 compared with earlier frontier models is also increasing uncertainty about Anthropic, which may be heading for an IPO as early as soon.

If this situation continues, business models at frontier AI model developers that have invested billions of dollars in training larger, more sophisticated models could also change significantly.

Miles Clements (마일스 클레멘츠), a partner at venture capital firm Accel, which has invested in Anthropic, told the FT, "Most people don't need to work at the frontier," and "The era when customers chose only frontier models was not sustainable."

He added, "Breakthrough technological advances are still essential to delivering the most ambitious promises AI companies have made, such as treating disease, and to attracting top research talent, but going forward these achievements will increasingly serve as a 'showcase.'"

The recent situation also appears linked to the rapid evolution of open-weights models that can be downloaded for free and modified to fit users' needs.

ㆍ[Tech Inside] Everything about open-weights models shaking up the AI landscape

Open-weights models have so far been led by Chinese AI companies such as DeepSeek, Alibaba, Z.ai and Moonshot, but the current situation is different. Even in the United States, Anthropic's home ground, industry moves toward open-weights are accelerating.

Nvidia's moves have recently drawn attention. The Information reported Nvidia is also developing an open-source-based large language model, Nemotron 4, with at least 1 trillion parameters.

At roughly twice the parameter count of Nemotron 3 Ultra unveiled in June, Nvidia's model may not be the world's largest open-source model, but it is seen as likely to rank among the top tier. The total parameter count of Kimi K3, a large language model recently released by Chinese AI startup Moonshot AI, is 280 trillion, and DeepSeek V4-Pro has 160 trillion.

ㆍ[Tech Inside] Why is Nvidia trying to develop large AI models itself?

Nvidia recently signed a $6 billion licensing deal with coding AI startup Poolside. It also includes hiring most of Poolside's engineers. Nvidia is also investing $1 billion, valuing Poolside at $12 billion pre-money.

The Wall Street Journal reported that Nvidia will use Poolside to speed development of a world-class open-weights model that can compete with Chinese AI companies such as DeepSeek and Kimi K3. The WSJ, citing sources familiar with the matter, said open-weights models have lower operating costs and are easier to customise, putting them in a position to directly challenge closed-model players such as OpenAI and Anthropic.

ㆍNvidia accelerates development of world-best open-weights model to compete with China