Remember when ChatGPT was a novelty phone app we typed simple, time-wasting questions into, satisfying our curiosity, and nothing much more? Yeah – those days are long gone. Today, AI services like ChatGPT and Claude (handled by OpenAI and Anthropic, respectively), can work through your files, and spend a worrying amount of time operating ever-growing major aspects of other apps and services on your behalf.
Somehow, despite all this extra functionality and complexity, AI companies are still trying to sell all this through the same simple choice – use the free version or pay a flat monthly fee.
That obviously isn’t going to last forever. But the question isn’t really what happens when OpenAI and Anthropic start charging for AI, because they already do. It’s which bits they’ll charge more for – and how much useful stuff – if any – will remain free.
The obvious answer is that subscription prices will rise. Maybe they will. But, ever the pessimist, I suspect it’ll be messier than that…
How much do ChatGPT and Claude currently cost?
OpenAI and Anthropic have arrived at very similar selection of subscriptions. At the time of writing, ChatGPT Plus and Claude Pro both cost $20 a month. Things escalate rather rapidly from there.
ChatGPT Pro has $100 and $200 tiers, offering five or 20 times as much usage. At the time of writing, however, OpenAI has temporarily paused new sign-ups and upgrades to its $200 plan. Existing subscriptions are unaffected, and the $100 tier remains available.
Claude Max uses the same $100 and $200 price points, again promising five or 20 times the usage of Claude Pro.
Very few people need either of those top-end plans, mind. If you occasionally use a chatbot to summarise an email or settle an argument about who played somebody in a film, paying $200 every month would be… well, not unhinged, per se… but it would certainly make your financial advisor’s eyebrows furrow in concern.
For a developer, researcher, or business that can turn all that AI work into glorious cash though, the higher tiers make more sense. Even these expensive subscriptions still have limits, though. Ten times the price mostly buys you more access, rather than an AI that’s ten times as clever.
How can they afford to give AI away for free?
Free services aren’t charity, although their users certainly appreciate the mercy shown to their wallets. The reality, though, is that hardly anyone is going to develop a daily ChatGPT habit if the first screen they see asks for a credit card. And with so many alternatives available, making a free service completely useless would simply send people elsewhere.
That’s very likely why ChatGPT’s Free tier includes unlimited text chats with its smaller GPT-5.6 Luna model. More demanding features, like deep research, remain limited.
OpenAI has also introduced adverts for Free and Go users in the UK. It says the adverts are separate from ChatGPT’s answers, and advertisers can’t see your conversations. You can opt out of them on the Free plan, but you’ll receive fewer daily messages in exchange. Surprised? Me neither.
Anthropic has actually made a point of going the other way. The company says that Claude will remain advert-free, arguing that adverts would create incentives that don’t always align with giving users the most helpful answer. Now that’s more my vibe.
That doesn’t mean Claude has discovered a way to run for free, mind. Anthropic relies on paid users and businesses, and has also mentioned introducing cheaper plans in some markets. Companies can (and have) changed their tune countless times in the past, so excuse me while I remain only tentatively optimistic.
AI prompts can massively vary in complexity
Something else to consider is the fact that flat-rate subscriptions have somewhat of an awkward problem – asking an AI to correct a typo, and asking it to examine 200 pages of documents are both technically one prompt. But they clearly aren’t equal jobs.
The same problem, incidentally, came up when I looked at how much water and electricity AI actually uses. A short text response hardly uses any power. Video generation and longer agent tasks are a different matter entirely.
In fact, OpenAI executive Nick Turley has compared unlimited AI subscriptions with selling an unlimited electricity plan. Now, that sounds very dramatic if you’re asking for a limerick about a hamster, but rather less so when an AI is beavering away independently for an hour, on some convoluted, godforsaken task that would turn your brain, with its weak reliance on “sleep” and “food”, to mush. No offence.
Charging for every individual prompt would be confusing and well beyond annoying though, so I doubt most consumer chatbots will suddenly sprout an electricity-style pay-as-you-go meter. Imagine! For what it’s worth, I think that bundles are more likely – a subscription could include plenty of ordinary conversations, but only a set allowance for heavier tasks. Once that’s gone, you might have to wait for it to reset or pay for more.
Parts of this are already happening, by the way. The labels just tend to say “usage limits” rather than “you’ve consumed this month’s allocation of expensive computer thinking, you silly goose”.
AI could still get cheaper
The thought of things getting cheaper in an age where streaming services like Netflix seemingly raise their prices 15 times a year seems utterly alien. But there’s a sliver of hope. Maybe.
In July, OpenAI cut the API price of its smaller GPT-5.6 Luna model by 80 per cent and reduced the price of GPT-5.6 Terra by 20 per cent. Those are developer prices rather than ChatGPT subscription fees, but it’s still a sizeable drop.
Smaller models can now do work that previously needed much larger ones, too. There’s also plenty of competition – if Claude becomes too expensive, OpenAI has a reason to undercut it, and vice versa. Yay!
Ultimately, basic AI may become cheaper at the same time as the best AI becomes more expensive. Short summaries and rewrites are easy to keep giving away, while an agent that spends half the morning completing a “real” task is much easier to sell as a premium service.
AI might remain cheap to talk to, then. Asking it to actually do things is where the bill starts growing. It seems fair, I suppose.
Which AI service should you pay for?
Personally, I wouldn’t subscribe to several AI services just because each one occasionally wins a comparison test. I’d stay with the free versions until their restrictions started regularly getting in the way of something useful, then pay for whichever one best suited that work.
There’s nothing wrong with subscribing for a particularly busy month and cancelling afterwards, either – aka, the Great Streaming Service Shuffle. If an AI service is saving hours on paid work every week, that’s also a decent argument for asking an employer or client to cover it as professional software.
Most people, though, probably won’t experience more expensive AI as a single, enormous increase. A new model might appear on a higher plan, or a useful tool might gain a monthly allowance. Each change will look fairly small on its own, and the subscription price might not even change at first. It’s a tale as old as time…
Read the full article here
