AI Original article:Claude Subscription deep dive:

AT A GLANCE

The report states that SemiAnalysis testing found that Claude some subscriptions'API equivalent value is relatively high, but quota rules can be adjusted in the backend.

Article

AI Subscription deep dive:Claude Subsidies even more generous than domestic Coding Plans

Beating AI Breaking news: SemiAnalysis tested the AI subscription plans of OpenAI, Anthropic, Meta, SpaceXAI, MiniMax, Moonshot AI, Zhipu, Cursor, and Cognition one by one. It tested how much subscription quota is consumed by input, cache writes, cache reads, and output, then calculated usage based on its own September Agent usage profile, in which approximately 96.6%is cache reading. It then applied each provider's API list prices to see how much the same batch of token would cost if it went directly through API the API. SemiAnalysis calls this figure the "API equivalent value."

Claude Anthropic's Sonnet and Opus receive the heaviest subsidies. With the same monthly 200 USD Max 20x plan, using Sonnet 5.5 can yield approximately 12,529 USD of API usage at most, equivalent to 62.6 times the monthly fee; Opus 5.5 is approximately 11,726 USD, equivalent to 58.6 times. Switching to the more expensive flagship model Fable 5.1 leaves only 2,485 USD, or 12.4 times. The same plan can differ by nearly 5 times because the rate at which different models and different types of token consume backend quota varies. Anthropic allocates substantial quota to Sonnet and Opus, while clearly tightening it for Fable.

Domestic Coding Plans also heavily subsidize heavy users, but the extent still falls short of Claude SemiAnalysis measured that 132 USD MiniMax Token Ultra is equivalent to approximately 3,088 USD MiniMax-M3 API in usage, equivalent to 23.4 times the monthly fee; 168 USD Zhipu's GLM Coding Max is approximately 11.6 times; 199 USD Kimi Code Ultra is approximately 6.7 times. Even for MiniMax, the highest among them, each USD of equivalent value is only about Claude of Sonnet and Opus.

OpenAI has recently begun actively tightening its subsidies instead. For newly purchased Pro 200 plans, the token quota for each model tier is directly halved; plans purchased previously can temporarily retain their old quotas until October 29. Now 200 USD ChatGPT Pro using GPT-6.1 Sol has an equivalent value of approximately 2,084 USD, equivalent to 10.4 times the monthly fee; at the same price,Claude Opus is 5.6 times as much. Even setting aside the difference in the two models'API unit prices and counting only actual token usage, the former is approximately 28600000000 more than 10200000000，Claude by 2.8 times.

But for the two companies' most flagship models, the gap is instead very small. 200 USD The GPT-6 plan uses 2,897 USD；Claude Fable 5.1 Astra, with an equivalent value of approximately 2,485 USD about Claude However, Fable has a separate quota limit and can consume at most half of the 2,485 USD plan's total quota. After using this

Key points

SemiAnalysis tested the AI subscription plans of OpenAI, Anthropic, Meta, SpaceXAI, MiniMax, Moonshot AI, Zhipu, Cursor, and Cognition.
The tests separately tracked subscription quota consumed by input, cache writes, cache reads, and output, and reported that in its September Agent usage structure, approximately 96.6%is cache reading.
The article states that the monthly fee 200 USD's Claude Max 20x when using Sonnet 5.5,API has an equivalent value of up to approximately 12,529 USD, which is 62.6 times the monthly fee.
The article states that 132 USD's MiniMax Token Ultra can be converted into approximately 3,088 USD's MiniMax-M3 API usage, which is 23.4 times the monthly fee.
The article states that the quotas for all model tiers in newly purchased OpenAI Pro 200 token have been halved; previously purchased plans can temporarily retain their old quotas until October 29。
The article states that, 200 USD's ChatGPT Pro when using GPT-6.1 Sol,API has an equivalent value of approximately 2,084 USD, which is 10.4 times the monthly fee.
SemiAnalysis estimates that the subscription business contributes approximately 10%in revenue to Anthropic, yet may consume more than 40%of inference compute capacity.
SemiAnalysis states that, among identical plans, one account had a quota approximately 20%lower; the relevant provider confirmed that the account had entered a very small-scale quota A/B test.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

“API“Equivalent value” converts the quantity of token available under a subscription plan into a monetary amount based on the API list price for the same model. On this basis, the report concludes that certain Claude plans provide heavy users with available usage exceeding the monthly fee paid, as well as that of some domestic Coding Plan and ChatGPT Pro plans compared in the article. However, this conversion depends on the tester's usage mix, model pricing, and backend charging rules; it does not represent cash actually received by users or the amount actually subsidized by providers.

Why it matters to readers

If the test results and rules are accurate, heavy users selecting AI subscriptions should not compare monthly fees alone; they should also consider model limits, the charging rates of different token, and quota changes. If providers adjust charging ratios in the backend, the actual available usage under the same plan may change.

Beginners can understand subscriptions as monthly plans with dynamic usage rules: a fixed list price does not mean a fixed quantity of model calls every month. Before choosing, confirm whether commonly used models have separate limits, and avoid judging value solely on “API equivalent value.”

Risks and unknowns

The main text is a restatement of SemiAnalysis testing and estimates and does not provide the original report, complete sample, or calculation process.
The article does not provide complete announcements from the relevant providers regarding quotas for all plans, model names, and adjustment rules.

AI Original article:Claude Subscription deep dive:

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center