Anysphere released Cursor 2.0 this morning, and the headline feature isn’t a new model — it’s how the editor gets out of your way. Native agent mode, previously a beta sidebar, is now first-class: type a multi-file refactor in plain English and Cursor opens a new pane that plans, edits, runs tests, and reports back when it’s done. We’ve been on the early-access build for two weeks. Here’s what changed and what didn’t.
What’s new in 2.0
The agent runs in a sandboxed terminal alongside your editor. It can read your codebase, write to any file you’ve granted access to, run shell commands, and now — new in 2.0 — pause to ask clarifying questions when stuck. Previously it would either guess or stop. The new behaviour feels closer to pairing with a junior who actually checks in.
The tab model has been rewritten. Anysphere claims a 30% latency reduction on multi-line completions; in our testing it was closer to 22%, but more importantly the cancel-mid-suggestion behaviour stopped feeling jittery. Small detail, big quality-of-life difference.
Pricing changes
The $20/mo Hobby tier is unchanged. Pro moves from $20 to $40/mo but now includes 500 fast Claude 4.7 Opus calls and unlimited Sonnet — previously you paid for Opus separately. For anyone using Cursor daily, this is a price cut disguised as an increase. Business stays at $40/user/mo with privacy mode enforced.
What didn’t change
The fundamentals: codebase indexing, the model picker (Claude / GPT-5 / Gemini), VS Code extension compatibility. If you didn’t like Cursor before, 2.0 isn’t going to convert you. If you already use it, this is the upgrade that closes the gap with terminal-native agents like Claude Code on autonomous tasks while keeping the GUI advantage.
The update rolls out automatically over the next 48 hours.
