We are gathered here today to mourn the passing of predictability. The simple, flat-rate subscription for AI coding assistants, a concept we held dear, has been laid to rest by Cursor. The AI-native editor's new pricing structure is a masterclass in financial engineering, a complicated will that carves up its user base into four distinct classes of inheritors.
What we've lost is the ability to just use the tool. I remember a time when my primary concern was whether the AI understood my prompt, not whether the prompt itself would tip me into a new billing dimension. That era of blissful ignorance is over. Now, every developer using Cursor must become a part-time accountant, weighing each "Chat with selection" against a mysterious, invisible meter.
The cause of death was a multi-tiered complication. Cursor's estate is now divided as follows:
Comparison data
| Price per month | Usage vs Pro | |
|---|---|---|
| Hobby | $0 | limited requests |
| Pro | $20 | 1x (baseline) |
| Pro+ | $60 | 3x |
| Ultra | $200 | 20x |
Hobby is enough to get a taste, but not enough to build a project. Pro is the new baseline, with frontier models from OpenAI, Anthropic, and Google, and it defines the "1x" usage unit. Pro+ costs three times as much for three times the usage; the math is, for now, simple. Ultra is the power-user tier: ten times the price of Pro buys a 20x usage multiplier and "priority access" to features yet to be named.
For teams, the math gets even more interesting. The Teams plan costs $40 per user per month. It includes the Pro features plus collaboration tools and SSO. I checked the numbers: that's a clean 100% markup over the individual Pro plan. For comparison, GitHub Copilot Business asks for a 90% premium over its individual plan. Cursor decided to round up.
The will's fine print is where the real inheritance is hidden. The pricing page cheerfully mentions usage multipliers but is silent on what they multiply. How many agent requests are in the Pro plan's "1x" baseline? How is "usage" measured: tokens, requests, CPU seconds? The document is silent.
It only promises that once you exceed your unspecified included amount, you'll be billed for overages in arrears. There's no free trial for any paid tier; you commit $20, $60, or $200 a month on faith alone.
And in a final, quiet clause, privacy is not a default right. It's a mode you must find and enable in the settings. If you don't, your code data may be used for model training by the providers.
So who inherits what? Hobbyists get a functional demo. Professionals get a $20 monthly bill with the potential for surprise overages. Teams get to pay double for central billing.
And a select few "power users" can pay $200 a month for the privilege of not having to think about usage, a luxury that was once, not so long ago, the default for everyone. The simple subscription is gone. Its replacement requires a spreadsheet.