Cursor to Users: Sorry You’re Confused, Now Pay Up Anyway

Anysphere, the company behind the AI coding assistant Cursor, has issued a public apology — not for overcharging users, mind you, but for not making it obvious enough that they were about to get nickel-and-dimed into oblivion.

In a blog post released Friday, Anysphere CEO Michael Truell extended his sincerest corporate regrets for a June pricing update that left Pro plan users shell-shocked, blindsided, and scrambling to check their credit card statements.

“We recognize that we didn’t handle this pricing rollout well and we’re sorry,” Truell wrote, in what appears to be the closest thing to accountability Silicon Valley is currently capable of mustering. “Our communication was not clear enough and came as a surprise to many of you.”

Translation: We changed everything, didn’t tell you properly, and now you’re mad.

Previously, Cursor’s $20-a-month Pro plan offered users 500 fast responses from cutting-edge AI models by OpenAI, Anthropic, and Google, with unlimited slower responses once the limit was hit. But with the June 16 update, the gloves came off. That same $20 now gives users access to just $20 worth of compute, based on the actual API rates charged by the AI model providers. And once you blow through that? Get ready to pull out your wallet — again.

The update also came with a charming little twist: unless users manually set a spending limit, they could keep making requests and rack up additional charges without any warning. Several users quickly found themselves hitting usage caps within a few prompts — especially when using Anthropic’s Claude models, which are notorious for their verbosity and appetite for tokens.

And, in a move only a SaaS platform could dream up, the only mode still offering unlimited usage is Cursor’s “auto mode,” which handpicks models based on what’s available — a roulette wheel of AI performance masquerading as a feature.

Truell explained the change as necessary due to the ballooning costs of newer AI models, which “spend more tokens per request on longer-horizon tasks.” In plain English: the models got more expensive, and Anysphere got tired of footing the bill.

Cursor’s users, of course, turned to the traditional battleground of discontent — social media. There, they shared screenshots of surprise charges, shredded the new plan’s lack of transparency, and questioned whether “Pro” still meant anything other than “Pay Ridiculously Often.”

Anysphere has since promised refunds to users who were unintentionally charged, and pledged to be “more clear” about future pricing updates. TechCrunch reached out for further comment, but the company declined to go beyond the blog post, presumably because the PR team is still recovering from the backlash.

The AI pricing mess isn’t exclusive to Cursor. Replit, another popular AI coding platform, recently triggered its own round of user fury after altering its pricing structure to make large AI tasks more expensive. The trend is clear: AI isn’t cheap anymore, and those costs are increasingly being dumped on users like unwanted software updates.

Anthropic’s Claude Opus 4 — one of the high-end models Cursor taps — runs at $15 per million input tokens and $75 per million output tokens. That’s higher than Google’s own most expensive model, Gemini 2.5 Pro, released in April 2025.

Meanwhile, Cursor’s competitors aren’t just content to overcharge; they’re actively luring away its customers. Anthropic’s Claude Code, an enterprise-targeted AI coding tool, has gained traction rapidly, reportedly helping push the company’s ARR to $4 billion. Not to be outdone, Cursor poached two Claude Code developers last week in a desperate attempt to stay competitive.

Cursor itself now boasts $500 million in annual recurring revenue, mostly from its Pro plan — a number it would very much like to keep climbing, even as it figures out how to offer “affordable” AI while outsourcing its brains to companies with no such incentive.

To keep ahead, Cursor is rolling out a $200-a-month Ultra plan, which includes generous rate limits and multi-year deals with OpenAI, Anthropic, Google, and Elon Musk’s xAI. It’s a bold move that suggests Cursor knows exactly who holds the compute power in this relationship — and it’s not Cursor.

Anysphere may still be apologizing, but the direction is clear: the future of AI coding is here, and it’s pay-to-play — with the user always paying more.