All tools
AI coding tools

Cursor

By Anysphere

AI-first code editor forked from VS Code, with deep autocomplete and an in-editor chat that knows your repo.

Visit CursorFreemium

Overview

Cursor is a VS Code fork from Anysphere that turned "AI in your editor" from a side panel into the main event. The fork model is the secret: by owning the editor, Cursor can ship deep agent features (multi-file edits, terminal control, codebase-wide indexing, an "agent mode" that runs tasks for you) without waiting for VS Code's plugin API to catch up. As of 2026 it's the default heavy-duty AI IDE for most full-time engineers. The model picker is a real differentiator. You bring your own choice of frontier model — GPT-5, Claude Opus 4.7, Gemini 2.5, plus Cursor's own optimized "fast" models — and switch per task. Tab completion is uncannily good at predicting your next edit (not just your next token), and Composer/Agent handles the multi-file refactors that used to take an afternoon. The honest tradeoff is cost and lock-in. At $20/mo Pro you'll burn through "fast" requests on any serious week and start hitting slow-pool waits, which pushes power users to higher-volume plans or the per-token pricing on Business. And once your muscle memory lives in Cursor, going back to vanilla VS Code feels broken. If you want a free-or-cheap autocomplete, Copilot is fine. If you want the full agentic IDE, this is it.

Best for

  • daily software development
  • large codebases
  • pair programming

Strengths

  • Tab completion predicts your next edit, not just your next token — feels like the editor reads your mind
  • Agent mode handles real multi-file refactors with terminal access and test-running
  • Model picker lets you switch between GPT-5, Claude, and Gemini per task without leaving the editor
  • Codebase indexing means @-mentions actually pull the right context across large repos
  • VS Code fork means every extension you already use just works

Weaknesses

  • Pro's fast-request quota runs out fast on heavy weeks; slow pool can mean noticeable latency
  • Closed source — you depend on Anysphere for pricing and roadmap
  • Indexing large monorepos is memory-hungry and occasionally goes stale
  • Privacy mode is opt-in on Pro — defaults send your code to Anysphere

Pricing

Hobby

Free

Limited "fast" requests per month, 2,000 completions, and access to a smaller default model. Enough to evaluate the product, not enough for daily professional use.

Pro

$20/mo

Unlimited completions, ~500 fast requests/month across frontier models, and unlimited slow requests. The default tier for individual developers.

Business

$40/seat/mo

Pro plus org-level admin, SSO, privacy mode (no training on your code), centralized billing, and usage analytics. SOC 2.

Enterprise

Custom

Everything in Business plus higher rate limits, dedicated support, custom contracts, and procurement-friendly terms.

Use cases

  • Full-time engineering on a non-trivial codebase

    Codebase-wide context, agent mode, and tab completion compound — the more your repo is indexed, the more leverage you get per keystroke.

  • Multi-file refactors and migrations

    Composer can plan the change, edit ten files coherently, and run your tests. Manually doing this is the work it replaces.

  • Working across unfamiliar codebases

    Ask the agent "where does auth happen here?" and get an actual tour of the relevant files. Faster than grep when the naming is bad.

  • Pair-programming with model choice

    Claude for thinking, GPT-5 for code-heavy tasks, Cursor-fast for autocomplete. Picking per-task is genuinely useful.

  • Prototyping a new feature end to end

    Agent mode will scaffold routes, components, and tests in one pass. Faster than typing it; messier than writing it carefully.

  • Onboarding a junior engineer

    They can ask the codebase questions instead of interrupting senior engineers for context. Real productivity unlock for new hires.

When not to use

  • You're happy with autocomplete-only — Copilot at $10/mo does that better per dollar
  • You're working in a regulated environment that forbids sending code to third parties (use a self-hosted alternative)
  • You prefer a terminal-first workflow — Claude Code or Aider fit better
  • You want fully open-source tooling — Cursor is closed source

Alternatives

See it compared

Glossary terms to know

Related skills

Other AI coding tools