By admin 
The rise of artificial intelligence technology a la OpenAI, Anthropic, and Google has come by many names, but one of these is new: “worse.”
Read more Possible deal on Iran divides US lawmakers largely along party lines
That is what users are saying about AI chatbots at a time where both the U.S. stock market is turbo bullish on the AI industry and fast-growing and unprofitable AI labs prepare for potentially blockbuster IPOs.
It’s not an isolated incident either. It appears to be sweeping the industry, with potentially massive ramifications for adoption.
Anthropic burns its goodwill
Anthropic, in many ways, brandished itself as a firm with a superior moral framework by standing up to the U.S. Government and setting limits on how its AI models could be used. Of course, much of that goodwill was completely burned by recent controversies with the company.
Charging more for using third-parties
First, the company had a billing controversy where it charged users “per-token” for using popular third-party tools like OpenClaw. This resulted in much higher bills, as well as worries about privacy.
Claude got dumber
Not long thereafter, users began to complain that Claude felt dumber, something which the company ultimately confirmed after weeks of pushback. Evidently, the company downgraded reasoning effort and had a series of other issues which made their model dumber, but they were conveniently changed in time for the launch of competitor OpenAI’s GPT-5.5 model.
Pulling the rug on coders?
The outrage continued after it made changes on its website that suggested that its Claude Code product would no longer be available for its Pro Users, something it later said it was “testing a change” for. More recently, Claude changed the estimates for its “average cost per developer per active day”, more than doubling them in a quiet and underhanded way.
Anthropic stealth-tweaked their claude code docs to increase the estimated average that a developer would spend a day from $6 to $13, and the “average for 90% of people” from $12 to $30 a day. Very nice stuff!https://t.co/9zHZMVpDrdhttps://t.co/xJz9kNLyfnpic.twitter.com/0cT4KTfauQ
— Ed Zitron (@edzitron) April 27, 2026
Notably, the recent changes in Claude’s prices have forced some well-resourced firms to reconsider its usage. On Thursday, a report said Microsoft had pulled their internal code licenses because of the pricing, even though coders internally preferred it to Microsoft’s own competing CoPilot product. If a $3 trillion company can’t afford it, then who can?
Google nerfs AI Pro, Ultra
Fresh on the heels of its Google I/O presentation, Google announced a series of new products, unveiled a controversial change to its search engine to make it ‘AI-first’, and touted a recent price reduction for its AI plans. But customers received an email on Wednesday with some bad news: your usage will be capped.
New “compute-based usage limits” arrive in Gemini
Effective immediately, Google announced that Gemini app users would face “compute-based usage limits.” In other words, if you ask Gemini too much, they’ll lock you out for five hours, assuming you haven’t somehow hit a fairly unknown weekly limit. One consolation for AI Pro subscribers is that they get a “4x higher usage limit than non-subscribers.” AI Ultra consumers get even more than that, but at a significantly higher price.
The usage limit is nebulous and not going to be well understood by most people, which will likely lead to confusion. Obviously, more complex thinking tasks and use of more advanced models will accelerate the pace at which users hit the usage limits, perhaps leading to frustration.
But the limits didn’t stop with Gemini. Google is also bringing compute-based limits to other AI products in their suite, like its AI filmmaking tool Flow and its AI coding tool Antigravity. They’ll now force you to buy AI credits for these features, with the courtesy 1,000 credits no longer included as a benefit of the base plan.
Users worry OpenAI is next
Unsurprisingly, users of OpenAI’s ChatGPT also worry that the company could be testing changes that might eventually water down its normal AI models, too. On X (formerly Twitter), one user posted a preview of an A/B test featuring a new prompt box with different “intelligence” options.
Read more Tiburones y cerveza: SeaWorld le está dando a la gente lo que quiere
So OpenAI cut the intelligence of their normal models (low to no web search)
Instant/medium/high are pure trash now
Only Pro works now pic.twitter.com/v1KPHOUiKD
— Zephyr (@zephyr_z9) May 22, 2026
Adam Fry, a product lead on ChatGPT’s Atlas browser, responded to a separate post, indicating that this is a small test that consolidates ChatGPT’s seven compute levels into the four most common. “Definitely not our intent to remove capability for our Pro users!” he said.
There were already worries mounting
The response to the post includes a mix of reactions, including many users asking to “opt out” of the changes. This comes as users speculate that changes have been made which make ChatGPT Pro slower. Most of that is anecdotal, but given the ongoing antics in the industry, it would not be surprising.
AI just got here (but it’s already getting worse)
In the early oughts of the large language model (LLM), users were free to generate and prompt to their heart’s content, providing a context library to train subsequent models. They were the product. Now that the models are becoming capable (and expensive), the rug pulling has begun.
Companies like OpenAI have sworn by the goal of making an artificial general intelligence (AGI) that is capable of self-learning and performing tasks on its own. Therein lies the promise of a utopian, automated AI-centric economy. But usage limits are evidence that it is just that; a pipe dream.
Proponents of the AI boom have said that this won’t be a problem. OpenAI CEO Sam Altman even said that the cost of intelligence will trend to zero as models, hardware, and energy become more efficient. However, it appears that trend has not taken yet with various shortages.
Is AI an Uber/Lyft situation?
There are no original ideas in tech venture capital. In the past, I’ve written about how AI companies have adopted a pricing model similar to another one-time VC obsession: rideshare companies like Uber and Lyft.
In the early days of these companies, institutional money helped to subsidize the price of rides, even to the extent the firms were woefully unprofitable. They put the taxicab industry out of business, then raised prices. This is hardly the first time we’ve seen this in business; it’s a tale as old as time.
Some call them “dumping prices”, others call it “predatory pricing.” Brands like Standard Oil, Walmart, and Microsoft were found guilty of the practices. It’s simple, cut prices, eliminate competition. But it’s more than that; it’s an altogether risky strategy.
Why dumping might not work for AI
Sure, they have a larger total addressable market (TAM) than rideshare; they have the whole labor economy in their crosshairs. But Uber and Lyft had cheap contract drivers. AI companies need to constantly refresh expensive hardware to build and train more sophisticated models.
Of course, the hope is that they get users hooked by running at a loss, then hike up prices to pay the electricity bill (and eventually, the hardware costs, too). We’re looking at the companies testing the levers and raising prices. With limits now arriving on many models, we might be watching the end of the AI subsidy.
It might be a slow, long death. However, if AI companies are ever to make money in any meaningful capacity, the subsidy must die. Whether businesses stick around, or choose to stand up their own in-house LLMs, is the multi-trillion-dollar question that will slowly mount.
The Arena Media Brands, LLC THESTREET is a registered trademark of TheStreet, Inc.
Read more Analysis-US Treasury rout tests Washington’s tolerance for higher borrowing costs
This story was originally published May 24, 2026 at 10:12 AM.