Companies Respond to Changes in AI Pricing Models with Separate Token Allocations

Kim, a 36-year-old male developer at an IT company, has recently been focusing on using Anthropic's generative artificial intelligence (AI) model, Claude. News that Claude's pricing plan would switch to a usage-based system prompted the company to allocate more usage units, called 'tokens,' to those who use Claude more. As a result, a workplace atmosphere has emerged where employees compete to use Claude more frequently to avoid losing their token allocations. Kim remarked, "For now, the company encourages us to use AI extensively without worrying about costs," but also added, "Compared to before, the amount of Claude tokens available to developers is noticeably decreasing."

AI companies are revising their existing flat-rate pricing models in order to address challenges such as the burden of computational resources and declining profitability. In response, domestic companies have begun exploring countermeasures, such as concentrating spending on areas that require more computational resources or allocating additional resources based on performance verification, to avoid paying excessive fees.

원본보기 아이콘

According to the IT industry and other sources on May 6, AI companies are transitioning from flat-rate models that allow unlimited service use for a fixed fee to usage-based pricing systems, or are considering such a move. On April 14 (local time), Anthropic switched its enterprise subscription plan, "Claude Enterprise," to a usage-based model. Customers using Claude Enterprise now pay a base fee of $20 per month (approximately 30,000 KRW) and incur additional charges depending on their usage of the generative AI model. Previously, customers could use a fixed number of tokens for a maximum of $200 per month. However, the Team plan, which targets companies with fewer than 150 users, does not implement usage-based billing.

Microsoft's open-source platform GitHub's coding tool "GitHub Copilot" will also transition to a usage-based pricing model starting next month on the 1st. The "Copilot Business" plan for small and medium-sized enterprises offers token usage equivalent to $19 per month, while the "Copilot Enterprise" plan for large companies provides $39 worth of tokens per month. Once the allocated tokens are used up, additional payments are required. GitHub Copilot enables rapid programming by automatically generating code based on the meaning of comments and function names. This pricing model change is expected to impact the coding industry that relies on AI tools.

There is also a shift toward usage-based pricing models among Korean companies. NHN Dooray, a subsidiary of NHN, has announced that it may change the pricing for its AI collaboration platform, "Dooray AI," to a usage-based system. If users add AI assistant (agent) functions to email, messenger, and project collaboration tools, they will pay according to their usage. NHN Dooray CEO Baek Changyeol stated during a press briefing held at NHN's Seongnam headquarters on April 28, "Until now, we have used a monthly subscription model, but with the launch of the AI agent, we will switch to a usage-based, quantitative plan."

원본보기 아이콘

AI companies are changing their pricing plans mainly due to frequent service disruptions and profitability issues. Usage surges have caused frequent access failures for Anthropic's Claude as the infrastructure has struggled to keep up. According to the Claude status page, the model's uptime from February 4 to May 4 was recorded at 98.71%. This marks a drop from the 99% to 100% uptime maintained from 2024 through January of this year, as a result of the spike in usage. Joe Binder, GitHub's Vice President of Product, also stated on the official blog on April 22, "Sessions running in parallel for extended periods consume enormous resources," and warned, "Without additional measures, the quality of service for all users will deteriorate."

The Wall Street Journal (WSJ) reported on April 27 that OpenAI's management is questioning whether they can sustain the massive expenditures on data centers amid stagnant weekly user numbers and sluggish revenue. OpenAI has been focusing on improving profitability, for example, by introducing advertisements to its low-cost generative AI model ChatGPT earlier this year.

Reuters Yonhap News

원본보기 아이콘

Korean companies that need AI tools are preparing for these pricing changes by adopting performance verification methods. For example, a game company referred to as "Company A" uses the "Claude Team Standard" plan, which costs $25 per person. If employees wish to upgrade to a higher-tier plan that provides more tokens and better performance, they must obtain company approval before making the purchase. Lee, a 40-year-old male employee at Company A, explained, "If a team runs out of tokens, they must clearly justify the need and demonstrate what outcomes can be achieved with a higher-tier plan in order to be allowed to use it."

There are concerns that, once usage-based pricing is fully implemented, it will be more difficult to use generative AI models as much as before. Computational resources may be concentrated on development positions that use many tokens, and tasks such as PDF file generation or summarization-which consume significant resources-may be discouraged. An industry insider said, "Not many companies have established guidelines for AI usage yet," but added, "It is a time when each company must consider more efficient ways to use AI."

© The Asia Business Daily(www.asiae.co.kr). All rights reserved.