SemiAnalysis, a semiconductor research firm, recently tested subscription plans from Anthropic, OpenAI, xAI, Cursor and other major AI providers, then converted the usable token volume in each plan into the price a user would pay for the same model through an API. The result: for some flagship models and plans, Anthropic’s Claude subscription delivers more than five times the practical usage value of OpenAI’s ChatGPT offering.
The comparison went beyond monthly fees
SemiAnalysis did not compare plans on subscription price alone. It calculated how much text users could input, how much content they could generate, and how much compute capacity was available through items such as cache read and cache write, then translated those figures using official API pricing.
That approach matters most for heavy users. For people who rely on AI every day for coding, research, data analysis or agent-based tasks, the real constraint is often not whether a model is included, but how many times it can be called in a day and how quickly usage caps appear.
ChatGPT Pro’s revised $200 plan offers less usable volume
Using Claude’s higher-end models as an example, SemiAnalysis said the subscription plan provides a substantial amount of usage. If a user bought the same token volume directly through the API, the implied value would be far above the subscription fee actually paid.
By contrast, the gap widened after OpenAI revised ChatGPT Pro usage limits. SemiAnalysis found that OpenAI’s new $200 monthly plan offers roughly half the usable token volume of the older plan across several models.
For developers, researchers and AI power users who had been using ChatGPT Pro heavily, that means the monthly fee did not fall, but the amount of model usage obtained per dollar dropped in a visible way.
The $500 tier appears to sell speed more than token volume
OpenAI has also introduced a higher-end $500 monthly plan with more allowance and access to GPT-6 Astra’s Ultrafast mode. OpenAI said Astra can reach output speeds of about 300 tokens per second in Codex, a setup aimed at heavy coding, agent execution and fast iterative workflows.
SemiAnalysis said, however, that the $500 plan does not increase total model usage in proportion to the higher price when compared with the earlier $200 plan. In practice, a large share of the extra spending goes toward speed rather than simply buying more tokens.
Claude and ChatGPT are being positioned differently
SemiAnalysis said the two companies are increasingly separating their product positioning. Users who care most about access volume for higher-end models may find Claude’s subscription plans more attractive. Users whose workflows depend heavily on real-time code generation, agent execution speed and the Codex ecosystem may still see value in ChatGPT’s premium tiers.
The data also points to a problem that is often overlooked: AI subscriptions cannot be judged by monthly price alone. Two plans that both cost $200 can differ sharply in token volume, model tier, rate limits and usage caps.
Heavy users are more likely to feel the gap
For general users, differences between $20 monthly plans may not be obvious. For professional users who spend long hours with AI every day, the value gap can widen quickly once they start hitting usage limits on a regular basis.
As AI shifts from a chat tool into part of software development, deep research, long-running agent tasks and automated workflows, the ability to keep using the strongest models is becoming one of the key competitive points in subscription plans.
Based on SemiAnalysis’ testing, Claude currently holds a clear edge if the comparison is limited to how much higher-end model usage a user gets for each dollar spent.
