Cursor apologizes for confusing Pro plan price changes

For $20 a month, Cursor users used to tap into top-tier AI models almost without thinking. Then, a few complex prompts later, their monthly limit was suddenly gone — and some found surprise charges on their credit cards.

Following weeks of backlash, Michael Truell, CEO of Anysphere (the startup behind Cursor), published a formal apology on Friday. Truell admitted the company botched the rollout and communication of a major pricing overhaul to its $20-per-month Pro tier.

From unlimited slow queries to API-rate limits

Under the old Pro plan, developers received 500 fast responses on flagship models from OpenAI, Anthropic, and Google, followed by unlimited slow-rate responses.

On June 16, Anysphere quietly changed the rules. Instead of that fixed pool, Pro subscribers were granted $20 worth of usage calculated at raw API rates. Once that balance hit zero, coding tasks paused unless users bought extra credits. Unlimited access was downgraded exclusively to an "auto mode," where Cursor picks models behind the scenes based on server capacity.

Programmers immediately flooded social media with complaints. Many noticed their balances evaporated after just a handful of prompts, especially when relying on Anthropic's new, coding-favorite Claude models. Others who never set spending caps were hit with surprise credit card charges. Anysphere has promised refunds for those accidentally billed and pledged clearer communication going forward.

Frontier intelligence isn't getting cheaper

Truell defended the pivot by pointing to reality: modern AI reasoning burns way more tokens on long, multi-step tasks. In short, advanced coding models got too expensive for Anysphere to subsidize.

While commodity models keep getting cheaper, frontier intelligence costs a fortune. Anthropic's new Claude Opus 4 runs $15 per million input tokens and a staggering $75 per million output tokens, surpassing Google's Gemini 2.5 Pro from April. On top of that, both Anthropic and OpenAI have started charging enterprise premiums for high-speed, priority access.

Cursor isn't alone here. Competitors like Replit recently blindsided users with similar price jumps on heavy tasks. But Cursor is also caught in an awkward dance with the very companies powering it. Anthropic recently launched its own Claude Code assistant, which reportedly helped push its ARR to $4 billion while siphoning away Cursor users. Cursor fired back last week by poaching two Anthropic engineers who led Claude Code's development.

To lock down supply, Anysphere recently signed multi-year deals with OpenAI, Anthropic, Google, and xAI, while rolling out a $200-a-month Cursor Ultra tier for power users.

Why it matters

The era of using frontier AI models as an all-you-can-eat buffet for $20 a month is officially over. If you rely on Cursor for heavy, autonomous coding with flagship models like Claude Opus 4, you now have to budget carefully and watch your spending caps. Developers who want flat-rate predictability will likely have to settle for auto mode or pay up for the new $200 Ultra tier.

This is also a major reality check for the entire AI wrapper ecosystem. Cursor is an undeniable hit, boasting over $500 million in ARR. Yet despite that massive scale, it remains uncomfortably vulnerable to the model providers upstream. When model labs hike API prices, add priority fees, and launch competing in-house tools like Claude Code, wrapper tools are squeezed from both ends. Until Cursor builds competitive in-house models, passing those soaring compute bills directly to end users appears to be its only realistic escape hatch.