Cursor

What it actually does

Cursor is a code editor forked from VS Code with AI built into the middle of it rather than bolted on the side. Your existing extensions, keybinds and settings carry over, so the switching cost is close to zero. Cmd+K does inline edits on a selection, the chat panel answers questions about files you have never opened, and Composer plans and applies changes across several files at once.

The part that justifies the money is codebase awareness. @-mention a file, a folder or a symbol and the model works from your actual repository instead of a snippet you pasted. It reaches frontier models from OpenAI, Anthropic and Google through one interface, and Privacy Mode exists for teams that need code kept out of training. G2 has it at 4.6 out of 5 across 315 reviews, and the agentic multi-file editing is the reason people keep paying while being angry about the billing.

What it costs

Free gives you 2,000 completions and 50 premium messages a month. Pro is $20 a month. Business is $40 per seat per month.

The catch is what $20 means now. On June 16, 2025 Cursor changed the model: the old 500 fast requests plus unlimited slow requests became "$20 of usage at API rates". Developers who changed nothing about their workflow started hitting "rate limit exceeded" or projected-overage warnings days into the month. Limits were never published in tokens or requests, so nobody could predict the cutoff. One user measured the real Pro ceiling at roughly 520M tokens, about $195 of usage, which means the value of your plan rests entirely on internal pricing you cannot see.

On-demand billing is the other trap. Threads describe surprise charges on a $20 plan with no warning email before the card was hit. Check that setting before you do anything else.

Where it falls down

Billing conduct. Users call the June 2025 change a bait and switch, "from unlimited to unusable", and the response did not undo it. The CEO apologised publicly in early July 2025, telling TechCrunch "We recognize that we didn't handle this pricing rollout well and we're sorry", and TechCrunch noted "The changes resulted in some users complaining that they unexpectedly faced additional costs." The apology covered the communication. The usage-based model itself stayed.

Refunds are where it gets worse. Annual subscribers report weeks with no human reply, only an AI support assistant repeating that they do not meet refund eligibility, including cases where the subscription had been cancelled and never used. Reports of that pattern run through 2025 and 2026. If you are considering a year up front, read that sentence again.

There is also a quality complaint that predates the pricing fight and outlived it: users report that output degrades under load or after model and limit changes, so the same prompt costs more and produces worse edits than it did earlier in the same subscription. And the tier ladder keeps growing, with Pro, Pro+ and Ultra, while limits tighten at the bottom rung. That is the shape users predicted and now report.

Who should buy it

Full-stack developers and small startup teams who write code every day should buy it, monthly, with on-demand billing switched off and a calendar reminder to check usage mid-month. It is the best agentic editor we can point at and the 3.5 reflects a product we rate highly and a vendor we do not trust with a stored card.

Do not buy the annual plan. Refund reports make that a one-way door.

Do not buy it if your AI coding needs are light. The free tier, or GitHub Copilot inside plain VS Code, covers autocomplete and occasional chat without exposure to a metered ceiling nobody will quantify for you. If you depend on niche VS Code extensions, test them first, because compatibility is good but not total.