Setup, Plans, and Model Selection

Installation and Migration

Download Cursor from cursor.com and install it like any desktop application. On first launch, Cursor offers to import your VS Code settings — accept this. It copies your extensions, themes, keybindings, and workspace preferences. You will feel at home within minutes.

If you want to run both VS Code and Cursor side by side during a trial period, that works fine — they are separate applications with separate settings stores. You can migrate fully when you're ready.

Plans and Pricing

Cursor moved to a credit-based billing model in mid-2025. The tiers as of 2026:

• Hobby (free) — limited usage per month, enough to evaluate Cursor but not for daily professional use.
• Pro ($20/month) — $20 in monthly credits. Covers roughly 225 Claude Sonnet requests or 500 lighter requests. Unlimited Tab completions. This is the right starting point for most developers.
• Pro+ ($60/month) — 3x the credit pool of Pro. For developers who run Agent frequently on complex tasks.
• Ultra ($200/month) — 20x Pro. ~4,500 Sonnet requests per month. For teams or individuals who use Agent as their primary development workflow all day.

The most important thing to understand about pricing: Auto mode does not draw from your credit pool. When you let Cursor pick the model, it uses a cost-efficient selection that is included in your plan. Credits are consumed only when you explicitly select a specific frontier model like Sonnet or Opus.

Model Selection: Which Brain for Which Task

Cursor makes multiple models available. Knowing which one to use saves both credits and time:

• cursor-small — Used automatically by Tab. You never select this manually. It is optimised purely for fast inline completions.
• Auto — Cursor's default: it picks the most cost-efficient model that can handle your request. Use this for all Chat and Agent work unless you have a reason to override it. Auto is included in your plan at no credit cost.
• Claude Sonnet — The best balance of quality and credit cost for most coding tasks. Switch to this explicitly when Auto's output feels shallow or when you're doing careful refactoring work.
• Claude Opus / Gemini Pro / GPT top-tier — For the most complex Agent tasks: large architectural changes, multi-step reasoning, debugging subtle bugs across many files. These consume credits quickly — reserve them for tasks that genuinely need them.
• Gemini Flash / lighter models — Fast and cheap for bulk operations: generating boilerplate, quick lookups, bulk renaming, low-stakes changes.

Codebase Indexing

When you open a project folder in Cursor for the first time, it automatically begins indexing your codebase. Cursor scans every file, builds a semantic graph of your code structure, and stores it locally. This is what makes @codebase search work — instead of searching by filename, it searches by meaning.

Indexing a medium-sized project takes a few minutes and runs in the background. You will see a progress indicator in the status bar. Once it's done, Cursor understands your project at a structural level, not just a text-search level.