According to external media, the open source AI model is rapidly expanding the use of the business, but this trend has not, for the time being, significantly compressed the revenue space of front-line model companies such as Anthropic. The business market is emerging with a clearer hierarchy: cheap models are responsible for large-scale production, and high-cost models continue to dominate early trials and complex tasks.

An AI executive recently suggested that once AI is mature, enterprises tend to shift some of their tasks to lighter and lower-cost models. At the same time, however, the overall expenditure of the front model has not decreased significantly. According to the article, this means that open-source models and front-line models are not necessarily simple substitutes, more like different stages in the same business chain.

Call to focus on open source models

From Vercel ' s AI Gateway data, DeepSeek has risen to the top of the token processing volume in the past week, accounting for more than one third of the platform ' s infrastructure traffic. The introduction of GLM-5.2 Z.ai also rose to the forefront during the same period.

OpenRouter data show a similar trend. DeepSeek V4 Flash, a week-by-week token scale of 5.3 trillion, became one of the most used models. By contrast, the forward model Opus 4.8 has slightly more than 2 trillion people per week.

This suggests that in high-frequency, standardized missions, enterprises are indeed increasingly willing to use cheaper models to reduce the cost of reasoning.

Anthropic still owns most of the expenses.

However, if measured by expenditure rather than by redeployment, the situation did not change by the same magnitude. According to Vercel data, Anthropic still accounts for more than half of the total AI expenditure of the platform and, although the share has declined slightly over the past month, it has not changed much.

OpenRouter does not directly rank total expenditure, but its price data reflect a similar structure. The average token cost of Opus 4.8 is about US$ 1.37 per million, while DeepSeek V4 Flash is about US$ 0.06, which is about 23 times the price of the latter. Based on this gap, it is likely that the front-line model will still take most of the revenues.

The article mentions that recent data are not even fully accounted for in Nvidia's Nemotron. Because of Nvidia ' s strong access capacity in the business market, the model could be followed by a rapid expansion of coverage.

Enterprise market or formation of a two-tiered structure

According to the article, the more rational explanation for this is that the number of tasks that AI can cover is still growing rapidly. Growing-up scenarios will gradually migrate to low-cost models, but new and complex scenarios are emerging and continue to provide demand for high-cost models.

Another explanation is that some enterprises still have more difficult tasks and cannot be completely replaced by cheap models for the time being. Thus, even if clients move part of the work stream to open source models, the front-line model will preserve the most valuable segment of the market, i.e. high unit price token and new scene validation.

In this judgement, the AI market may evolve into a relatively stable two-tiered structure: the front-line model is responsible for identifying new uses and the open-source model is responsible for large-scale deployment when mature. At least according to current data, the rise of open source models has not yet had a significant impact on Anthropic.