AI subscriptions have turned the monthly software bill into a scratch card. You know what you paid. You don't quite know how much work you bought.
That's not an accident. The labs are selling access to scarce computing power, but they price it like a gym membership. Most people won't use the full allowance, heavy users will discover the fire exit, and nobody wants to print a clean unit on the door.
The same staircase, twice
ChatGPT Plus costs $20 a month. Its two Pro tiers cost $100 and $200, with five times and twenty times more usage than Plus. Claude Pro is also $20. Claude Max then costs $100 for five times the Pro allowance and $200 for twenty times.
Two rival labs, the same three rungs. Apparently, innovation stops just before the pricing table.
The neat symmetry makes comparison look easy. It isn't. Twenty times what? It doesn't mean twenty times the useful work, twenty times the answers or twenty times the hours. With Claude, the allowance changes with message length, attached files, the length of the conversation, the model, the tools and the effort setting. Multi-step jobs use more too. OpenAI's Plus limits can shift with system conditions.
So the customer isn't buying a measured quantity. They're buying a larger chance of not being interrupted. That's a perfectly real benefit, but it's a rather different product from the one suggested by a tidy monthly price.
Unlimited, mostly
OpenAI calls part of Pro unlimited under its abuse guardrails, while access to its Pro reasoning model still has a usage limit. Claude Max resets its session allowance every five hours, adds a weekly limit across all models, and leaves room for other caps on models and features.
Unlimited has become an emotional support adjective. It sits on the pricing page and makes everyone feel better while the meter keeps doing cardio.
Claude also lets paid users carry on after the included allowance by buying usage credits at standard API rates. Customers can set a monthly spending cap, turn on automatic reloads or choose no spending limit. That's more honest than pretending the meter doesn't exist, but it changes the product again. The subscription gets you through the front door. Consumption billing is waiting in the corridor with a clipboard.
Call me cynical, but the labs have found a lovely arrangement. They get predictable subscription revenue. The customer gets an allowance that depends on what they ask, how they ask it, which model answers and how busy the machinery happens to be. Cui bono isn't a long meeting.
The only useful unit
For a business, the wrong question is whether one chatbot seat is cheaper than another. The useful question is what a completed job costs, and what happens when the allowance runs out.
A founder using AI to tidy a memo can wait for a reset. A company using it for research, customer triage or software work has a different problem. An interruption halfway through a real process isn't a minor inconvenience. It's part of the cost.
The sensible comparison is a representative week of actual work. Count the recurring jobs completed, where limits appeared, whether a cheaper model could do the dull bits, and what extra usage cost. Don't measure messages. A one-line prompt can start a large job, while a long prompt can produce something useless with admirable efficiency.
Some opacity is technically understandable. AI workloads genuinely vary. But variable cost is the reason to expose the meter, not hide it. If a platform can count every scrap of computing for billing, it can probably manage a better answer than "limits may vary". Apparently, arithmetic becomes experimental when it reaches the sales page.
My read is that the next useful fight in AI won't just be over which model is cleverest. It'll be over honest metering, hard caps, clear alerts and receipts a normal finance team can understand. Model quality wins the demo. Predictable cost wins the renewal.
AI has spent years promising intelligence. It can start by producing an intelligible bill.
Which company will show the meter before asking for the card?
Sources
- https://help.openai.com/en/articles/6950777-chatgpt-plus
- https://help.openai.com/en/articles/9793128-about-chatgpt-pro-plans
- https://support.claude.com/en/articles/8325606-what-is-the-pro-plan
- https://support.claude.com/en/articles/11049741-what-is-the-max-plan
- https://support.claude.com/en/articles/9797557-usage-limit-best-practices
- https://support.claude.com/en/articles/12429409-manage-usage-credits-for-paid-claude-plans