Premium AI subscriptions from OpenAI and Anthropic priced at about $200 a month were assessed to provide API usage value worth thousands of dollars when converted based on actual use. If heavy users who consume large volumes of tokens, such as coding agents, use their limits to the maximum, AI companies end up bearing substantial costs, critics say.

Cryptopolitan, a blockchain outlet, reported on Aug. 24 that research firm SemiAnalysis bought OpenAI and Anthropic’s premium subscription services and put them to long coding tasks. It estimated the Anthropic service to be equivalent to about $8,000 a month in API usage and the OpenAI service to about $14,000.

That is far higher than the market had expected. Some had projected that a $200-a-month subscription could provide token use worth about $2,000 a month, but using the service up to its limit can draw 3 to 7 times more API usage than that.

The two companies’ subscription structures are also similar. Anthropic’s Max plan starts at $100 a month and provides 5 times or 20 times more usage than Claude Pro. OpenAI’s $100-a-month plan provides 5 times the usage of Plus, and its $200-a-month plan provides about 20 times the usage.

It is not easy for ordinary users to actually consume that much usage. SemiAnalysis tested by repeatedly running very long coding tasks and exhausting weekly usage limits as much as possible. This means that even with the same $200-a-month subscription, the actual value and the costs borne by AI companies can vary greatly depending on usage patterns.

Token consumption is rising quickly, especially with the emergence of coding agents. SemiAnalysis assessed that agent-based coding can use about 1,000 times more tokens than simple coding questions. It noted that token usage can differ by as much as 30 times even for the same task, but said using more tokens does not necessarily improve the quality of the output.

There are also ways to reduce costs. Another study found that actively using prompt caching could lower API costs to about $0.57 per 1 million tokens, with a cost-reduction rate reaching 88.6 percent. But if heavy users consume API usage worth thousands of dollars with only a $200-a-month subscription, AI providers still face substantial cost burdens.

Another variable is that AI companies’ results are growing quickly. Anthropic’s second-quarter revenue surpassed $11.6 billion, more than doubling from a year earlier, and OpenAI’s revenue rose 18 percent to $6.7 billion. Anthropic’s annualized revenue run rate also exceeded $65 billion.

The problem is that rising AI usage may not translate directly into improved profitability. Ramp’s August AI index showed AI adoption rates among U.S. companies at 43.5 percent for Anthropic and 39.7 percent for OpenAI. But Anthropic’s most expensive model, Fable 5, accounted for only 6 percent of total tokens purchased by customers in its first month after launch.

There is also the possibility that inference costs will rise further. Gartner forecast that inference costs could increase by more than fivefold by 2028 as agent-based AI spreads. Even if token prices keep falling, the number of tokens processed by AI agents could rise sharply, pushing overall costs higher instead.

Ultimately, OpenAI and Anthropic face a choice: absorb the costs from maintaining high usage limits for premium subscriptions or adjust usage terms. With open-source AI models spreading on the back of price competitiveness, how sustainable the $200-a-month subscription can be as a profit model is emerging as a new variable.