OpenAI and Anthropic's $200 AI plan provides thousands of dollars in API equivalent usage

According to a study by research firm SemiAnalysis, the actual usage of two flat-rate AI subscription services that charge $200 per month seems to far exceed the value of the price. For OpenAI and Anthropic, this forces them to face the question: How long can subsidized technology access last when the market is putting pressure on AI profits?

According to SemiAnalysis, they purchased subscription services from both companies and used them to perform a lot of coding until they reached the maximum amount of work allowed per week. Based on standard API rates, Anthropic's $200 monthly Claude plan actual service costs approximately $8000. OpenAI's $200 plan costs approximately $14000. The study has been completed, preliminary results were released in June, and supplementary results were released on August 23.

Why the $2000 assumption has become five-digit

These numbers are much higher than many in the industry expected. According to SemiAnalysis, the industry had previously believed that a $200 subscription would generate approximately $2000 worth of tokens per month. But those users who make full use of their weekly quota could actually earn three to seven times that amount.

Anthropic's Max plan starts at $100 per month and offers 5 or 20 times the usage of Claude Pro. OpenAI adopts almost the same pricing structure, with its $100 plan offering 5 times the usage of Plus and its $200 plan offering 20 times the usage. Maximum API equivalent token value by AI subscription level.| Source: SemiAnalysis, June 2026. The chart shows the usage value calculated at standard API prices when the test limits were fully utilized, rather than the actual cost to the service provider of these tokens.

Most subscribers will likely never reach these caps. SemiAnalysis achieves this result by continuing to execute extremely long cycles of workload rather than regular conversations.

The token bill behind proxy coding

This is significant because coding agents often have high demand for the number of tokens. A study of eight cutting-edge models conducted on SWE-bench Verified shows that proxy encoding may use about 1000 times as much tokens as a simple single-round encoding problem, with input tokens accounting for the bulk of the cost.

Estimating usage proved very difficult. The same study found that the token consumption difference for tasks completed in the same way may be as much as 30 times, and that greater token investment does not necessarily lead to better results.

There are several ways to reduce costs. In another independent company study, heavy use of prompt caching reduced API costs to only approximately US$0.57 per million tokens, or even 88.6%. However, users who received $8000 and $14000 worth of API usage from a subscription plan that only cost $200 still received a considerable amount.

This positioning in the AI market

As competition among top laboratories intensifies, pressure is also increasing. The report shows Anthropic's second-quarter revenue exceeded $11.6 billion, more than doubling from the same period last year, while OpenAI's revenue increased 18% to $6.7 billion (according to a report). In addition, Anthropic's annualized revenue operating rate has exceeded US$65 billion.

Ramp's August AI index showed that Anthropic leads U.S. corporate adoption at 43.5%, compared with 39.7% for OpenAI. However, Ramp chief economist Ara Kharazian also pointed out that there is a limit to the amount companies are willing to spend: Anthropic's most expensive model, Fable 5, accounted for only 6% of total tokens purchased by customers in its first month of launch.

Gartner also predicts that by 2028, the emergence of proxy AI could increase reasoning costs more than fivefold. The company defines this situation as a "paradox of inference" that a drop in token prices does not mean a decrease in costs because agents will use more tokens.

This is why SemiAnalysis's data is important. If power users can get five-figure API equivalent usage from the $200 plan, OpenAI and Anthropic will eventually face the choice: absorb subsidies or tighten access. Either way, pressure on AI margins is likely to grow as cheaper and increasingly powerful open source models provide customers with more alternative options.