Claude Max can include $100 or $200 a month in API credits and still stop you using the assistant when you hit its limits. I don't think that's a scandal. I think it's a useful lesson in how AI companies sell the word "more".
A subscription allowance and a credit balance aren't the same thing. One lets you work inside a product. The other pays for a particular kind of usage elsewhere. Put them together on a pricing page and they look wonderfully generous. Try to spend one in place of the other and the generosity becomes rather specific.
My view is that the bundle makes most sense for people building with Claude, not simply people who want to talk to it for longer. That's a distinction worth making before anyone starts doing enthusiastic subtraction on the monthly fee.
More of what, exactly?
Claude is Anthropic's AI assistant. Its Pro plan costs $20 a month in the US, with regional prices and taxes varying. Max starts at $100 a month and offers options with five or twenty times Pro's usage. Those are bigger allowances, not an invitation to move in permanently.
Pro has a five-hour session limit and a weekly limit. The amount of work you get through depends on things including message length, attached files and the model or feature you're using. Claude and Claude Code, its coding tool, share usage limits on Pro and Max.
That last bit matters. Two tools don't necessarily mean two supplies. It's two straws in the same drink. Giving the second straw a professional-looking name doesn't refill the glass.
I don't object to limits. Computing costs money, and a fixed subscription needs boundaries. I object to treating a list of included tools as though each comes with its own independent allowance. For a business owner, the useful comparison isn't how many things appear in the bundle. It's how much of the actual job the allowance will cover.
A founder asking occasional questions and a developer handing over long coding tasks aren't consuming an identical service just because both have a chat box.
The credit isn't a refund
The other meter is the API. That's the connection software uses to send work to an AI model without someone sitting in the chat app. Think of an application sending customer enquiries for sorting, rather than a person pasting them in one at a time.
Max 5x includes $100 in monthly API credits; Max 20x includes $200. Eligible subscribers claim them by linking a Claude Console organisation, the account that manages this software usage. New subscribers need seven days on an eligible plan, and the offer is rolling out gradually.
Those credits don't increase the subscription's usage limits. They don't cover interactive Claude Code sessions or extra usage in the Claude apps. Unused credits expire at the end of the billing cycle.
It's an all-you-can-eat restaurant where the voucher's valid only in the kitchen. Excellent news if you've come to open a restaurant.
For someone already building an application on Claude, the credit can have genuine value. For someone who only wants a longer conversation with the assistant, it doesn't solve that problem. I wouldn't call a credit useless because it can't do something it wasn't designed to do. I also wouldn't count its face value as money saved unless there's eligible work to spend it on.
Otherwise, we're valuing a gym membership by the retail price of all the showers we could theoretically take.
Follow the second bill
Call me cynical, but I read this as more than a generous subscription perk. Anthropic benefits if a paying assistant user becomes a builder whose software also sends work to its platform. The credit makes that next step easier. It's a sample of the metered business, not cash tucked into the subscription box.
That's my read on the incentive, not evidence of anything concealed. The boundaries are there in the documentation.
The next boundary matters too. When the monthly API credit runs out, usage can continue against purchased credits or an enabled automatic top-up. Without other credits, requests stop. There's no need to invent a surprise-billing villain here. There's a perfectly ordinary business question: which supply is the work using, and what happens when it runs dry?
I think that question beats another argument about which model sounds cleverest. A tool can be impressive and still be awkward for a particular workload. A bundle can be good value and still include benefits a particular buyer won't use.
For founders comparing AI subscriptions, I'd judge the package against a defined piece of work, including the human checking, the shared allowance and any separate software bill. Not the number of features. Not the face value of a credit that'll expire unused. Certainly not the comforting size of the word "Max".
The interesting business here isn't just selling a better answer. It's turning the person asking questions into a customer with an application that keeps asking them.
The assistant gets you through the door. The meter gives you a reason to stay on the invoice.
Is the better business selling the assistant, or supplying everything its users go on to build?