As PCMag’s resident AI expert, I test everything from AI chatbots to AI music video generators daily. But no matter what I’m doing, I never really know how much usage I can get out of a given service before it cuts me off. How many images can I generate? How much coding can I do? The answers are just totally unclear. Below, I take you through the many ways AI companies obfuscate usage limits and what needs to change.
With AI Services, You Pay for Usage, Not Features
When you subscribe to any AI service, you’re essentially paying for temporary access to a data center with the hardware necessary to enable a given capability. AI usage, as a concept, simply refers to how much access you have to an AI product before you hit a specified limit. To continue using a service after you expend your allotment, you have to wait for a limit reset or spend more money to unlock more usage. AI companies use all kinds of terms to define AI usage, whether it’s credits, subscriptions, tokens, five-hour limits, weekly limits, and more.
For example, I pay for a Google Drive storage plan, which includes access to Gemini AI Pro. As an AI Pro subscriber, I get 1,000 Google Flow credits to use each month with Gemini’s excellent AI video generation models, Omni and Veo, among other things. This is a common way companies sell AI usage, but it’s far from the only one.
The only time you don’t have to pay for some form of access with an AI service is if you run it locally on your computer. Realistically, though, that’s often not an option. Most AI apps don’t let you run anything locally, and even if they did, you might not have the hardware necessary to ensure a good experience.
The ‘More’ Trap: Paying for Vague Tiers
I enjoy vibe coding apps to help me improve at video games, such as Magic: The Gathering Arena and Warframe. My favorite tool for those projects at the moment is ChatGPT, and I have a ChatGPT Pro 5x subscription that costs $100 per month. What is a Pro 5x subscription, you might ask? It’s one of ChatGPT Pro’s two flavors: 5x or 10x, and it refers to how much more usage you get than with ChatGPT Plus.
ChatGPT’s subscription plans don’t attach numbers to usage (Credit: OpenAI/PCMag)
How much usage do you get with ChatGPT Plus? Well, for starters, you get more than you do with ChatGPT Free or ChatGPT Go. Confused yet? OpenAI advertises Plus with nebulous descriptors, such as “expanded Codex usage” or “expanded messages and uploads,” but doesn’t attach concrete numbers to these claims. So, from the very beginning, you’re paying for the abstract idea of ‘more’ usage, not a certain amount. This makes it impossible to determine which of ChatGPT’s plans is right for you.
This kind of setup also means OpenAI never has to raise prices. Netflix makes headlines every time it hikes rates, but imagine if your Netflix subscription let you watch only an unknowable amount of programming each month, depending on the things you choose to watch. If you hit your limit in the middle of your favorite show, you would have to upgrade your subscription or pay for continued access. Then Netflix could just quietly change the monthly limits on each plan and subtly push you to spend more money without ever raising its prices. That’s how AI services currently work.
Moving the Goalposts With Random Resets and Hidden Caps
The situation goes from bad to worse once you actually start using an AI service, thanks to arbitrary rate resets. For example, since the launch of GPT-5.6, OpenAI reset weekly usage a bunch of times, letting you get another full week’s worth of access even if you had none left. This makes it really difficult to determine which ChatGPT plan will provide enough usage for your needs.
Beyond this, OpenAI is reportedly now testing paid usage resets. If that sounds a lot like the first hit being free and the rest costing you a pretty penny, well…
You can bank ChatGPT usage resets for later, but that won’t mean much if OpenAI starts selling them (Credit: OpenAI/PCMag)
Aside from usage resets, OpenAI obscures limits in other ways, too. For example, ChatGPT used to have both five-hour and weekly usage limits. OpenAI recently did away with the former, which makes it more convenient to use than Claude (which still has them). However, the new system still encourages you to quickly blow through your usage and then spend money to unlock more, upgrade to a more expensive plan, or wait for (or buy) a potential usage reset.
Discounts That Disguise the Real Cost
Promotions also add to the confusion. For example, as of this writing, Claude is running a temporary promotion in which you get 50% extra usage in Claude Code. OpenAI has run similar promotions, too. On a fun note, this promotion is for weekly limits only, not five-hour limits. In other words, you won’t be able to get more out of a single session. Instead, you need to use Claude for as many five-hour periods as possible during a week to take advantage. Considering how few five-hour periods you are likely able to sit at your computer and vibe code during a given week, it’s entirely possible that you might not benefit at all.
At face value, promotions are a win for consumers. But if you dig deeper, they end up being clever ways to hide how much usage you really need to complete a project. And if you have to upgrade eventually, that can mean a jump from $20 per month to $100 per month, or from $100 per month to $200 per month, depending on the plan. Also, when I say temporary promotion, I don’t mean something that lasts just a few weeks. Claude’s current promotion has been running since May 13, which is more than enough time to forget what you’re actually paying for.
Upgraded Models Mean Downgraded Value
The unending onslaught of new models and pricing adjustments further complicates this already puzzling situation. Setting aside whether you actually need the latest AI models (you don’t), fresh releases often mean higher usage rates. For example, when Claude launched Fable 5, its most powerful model ever at the time, it burned through usage significantly faster than Anthropic’s previous top model.
It’s one thing if, for example, you’re a gamer and you have to buy a new console every seven or so years. It’s another thing altogether if you pay for a subscription to a service to use its top-tier tech, but then the top-tier tech changes several times a year and the value of your subscription plummets. Now, you might think that you just don’t have to use new models, and that’s sometimes true, like with Fable 5. However, that’s not always an option.
Can a model released only a few months ago really be called “legacy”? (Credit: OpenAI/PCMag)
For example, the oldest model available to me on ChatGPT’s web interface is GPT-5.5, which OpenAI calls a legacy model. Mind you, this “legacy” model launched in April 2026, just under four months ago. So, if you really like the blend of performance and usage you get out of a particular model, you can only use it for so long.
Recommended by Our Editors
Stop Paying for the Newest AI Models. You Really Don’t Need Them for Most Tasks
ChatGPT vs. Claude: After Testing Both, I Have a Clear Favorite
10 Everyday AI Prompts That Could Get You Sued, Fired, or Flunked
New models sometimes have additional restrictions, too. Fable 5, for example, is available only to certain tiers of Claude subscribers. And even if you can use it, you can only do so with 50% of the weekly usage allocation of your plan. I have a Claude Max 5x plan, which costs $100 per month. How much can I use Fable 5? Well, I can use it for 50% of 5x the usage of Claude Pro, which gives me “at least 5x more usage per 5-hour session than [Claude] Free.”
Even if you don’t pay for AI model usage with a subscription and stick with the same model, you’re still not safe. DeepSeek, for instance, is well known not only for funneling foreign data directly to the Chinese Communist Party, but also for its low API access costs. API access for AI models is a pay-as-you-go system in which each prompt you send incurs processing costs, and DeepSeek charges you based on the number of tokens you use. Recently, DeepSeek massively hiked the cost of tokens for existing models. So, now you can either spend a lot more money or use DeepSeek a lot less than before.
How to Fix AI Pricing: Just Give Us the Raw Numbers
With some technology, variability is baked into the cake, and understandably so. For example, when you pay for internet access, the speeds associated with your plan are estimates because they can vary based on several factors. With AI, it’s inevitable that new models will come with different value propositions from older models, and that’s OK.
What’s not OK is how AI companies continue to rely on sketchy monetization practices. Instead, they could simply allocate a clearly defined number of tokens to each plan per month. And, there’s no reason why older models have to disappear so quickly, either, especially if they’re popular or more efficient. Yes, token allocations might need to change from time to time to accommodate new models and features, but every other company has to deal with that, too. Why should AI companies get a pass here?
I’m not asking to get more for less here, just to know what I’m actually paying for. I’m more willing to spend extra money on something when I don’t feel like it’s a form of gambling. As it stands, I’d rather downgrade to a cheaper plan (or temporarily sign up for one) to make the most of promotional periods and usage resets than spend on a more expensive plan.
You shouldn’t stop using AI just because the pricing is a mess. After all, I still use streaming services even though I hate how many I have to pay for to watch the content I want to see. Instead, vote with your wallet by downgrading when tiers get vague, cycling plans, and refusing to pay for overpriced top-tier black boxes. You don’t have to quit the tech entirely, but more than ever, you need to be diligent about staying on top of what you actually need—and what you’re willing to pay for.
About Our Expert
Ruben Circelli
Writer, Software
Experience
I’ve been writing about consumer technology and video games for over a decade at a variety of publications, including Destructoid, GamesRadar+, Lifewire, PCGamesN, Trusted Reviews, and What Hi-Fi?, among many others. At PCMag, I review AI and productivity software—everything from chatbots to to-do list apps. In my free time, I’m likely cooking something, playing a game, or tinkering with my computer.
Areas of Expertise
Latest By Ruben Circelli
- Switching to Claude? This Easy Trick Brings Your ChatGPT Memory With You
- I Spent Hours Testing NSFW AI Video Generators. These 5 Are the Best
- I Tested the Top NSFW AI Image Generators. These Are the 7 Best
- The Best AI Chatbots for 2026
- The 5 Best AI Tools to Maximize Your Online Meetings
- More from Ruben Circelli
