What is Cursor?

Cursor is an AI-powered code editor built by Anysphere, a San Francisco startup founded in 2022 by Michael Truell, Sualeh Asif, Arvid Lunnemark and Aman Sanger, with early backing from the OpenAI Startup Fund. It’s a fork of Visual Studio Code, so it looks and feels familiar and runs every VS Code extension, theme and keybinding — but the difference is architectural. Where most assistants sit inside the editor as a plugin, Cursor was rebuilt from the ground up around AI, giving the model deep, indexed awareness of your whole codebase.

That bet paid off spectacularly. By early 2026 Cursor had become the fastest-growing software product in history, reaching around $2 billion in annual recurring revenue — roughly double its figure from three months earlier — with more than two million users and over a million paying subscribers. It’s now the default professional choice for AI-native development, and its main rival is the incumbent it disrupted, GitHub Copilot. For the underlying model landscape that powers both, our guide to which AI model to use in 2026 is a useful companion.

Key features

Tab autocomplete — the standout

Cursor’s most loved feature is its prediction engine. Built on technology from the Supermaven team it acquired, Tab doesn’t just complete the current line — it predicts your next edit anywhere in the file, including multi-line refactors and changes several lines away from your cursor. Developers routinely describe it as “telepathic,” and it’s the single feature that converts people from other tools. For day-to-day flow, nothing else feels as fast.

Composer and multi-file editing

Composer is the killer capability for real work. Describe a change in plain language — “add authentication to these routes,” “rename this concept across the project” — and Cursor edits multiple files, or the entire repository, in one coordinated pass, showing a reviewable diff before anything is applied. It’s the difference between an assistant that helps you type and one that helps you build.

Agent mode and background agents

Agent mode runs autonomously: it plans an approach, writes the code, executes commands in the terminal, reads the output and fixes its own errors in a loop until the task is done. Cloud (background) agents take this further, running tasks in parallel on remote machines while you keep working — though these bill separately and require MAX mode, which adds a 20% surcharge on token costs. It’s powerful, and it’s also where the meter runs fastest.

Codebase intelligence and multi-model choice

Cursor indexes your whole project with @codebase so answers and edits reflect your actual architecture, conventions and dependencies rather than generic suggestions. It’s model-flexible too: you can run Claude Sonnet and Opus, GPT-5 and o-series models, Gemini 2.5 Pro or Cursor’s own fast model, and switch per task. It supports MCP natively for connecting external tools, includes an in-editor PR reviewer (the Bugbot reviewer is a separate paid add-on), offers a privacy mode that guarantees code isn’t stored remotely, and imports your existing VS Code setup in one click.

The standout: AI-native, not AI-added

Cursor’s real distinction is that the AI isn’t a feature — it’s the foundation. Because the editor was rebuilt around whole-project understanding, the combination of telepathic autocomplete, repo-wide Composer edits and a self-correcting Agent works as one coherent system rather than separate add-ons. That integration is why power users consistently rate it the most capable tool in the category, and why it overtook far older, better-funded incumbents in under two years.

Scorecard

Overall score: 8.5 / 10 — the average of the eight categories above.

Pricing

Cursor moved to a credit/usage-based model in mid-2025, and understanding it is essential before you subscribe. Each paid plan includes unlimited Tab autocomplete plus a monthly pool of credits that’s consumed by chat, Composer and agent requests — the heavier the model and the more autonomous the work, the faster the pool drains.

The honest read: Pro at $20 is the right starting point, but treat the credit pool as the real spec, not the headline price. For autocomplete-led work it’s excellent value, but if you lean on Composer and agents for frontier-model tasks, many users report burning through Pro credits and effectively paying $40–50/month after overages — or stepping up to Pro+ or Ultra. Cloud agents and MAX mode add a further 20% surcharge. An Enterprise tier with custom pricing adds SAML/SSO, audit logs and centralised billing. If that variability worries you, GitHub Copilot‘s lower entry price and unlimited free completions are the more predictable alternative — though it’s a plugin, not an AI-native editor.

Pros & cons

Honest weaknesses

Cursor’s biggest problem isn’t the product — it’s the billing story. In mid-2025 the company shifted from a clear request allowance to a credit-based usage model, and the change cut the effective number of premium requests at the $20 tier sharply, with little migration period and confusing communication. The backlash was serious enough that the CEO issued a public apology, and some developers moved to rivals. The model has since stabilised, but the core tension remains: the most powerful features — frontier models running in agent mode — are exactly the ones that drain credits fastest, so the people who get the most from Cursor are also the most exposed to surprise costs.

There’s a predictability gap beyond price, too. Output cost varies with which model you pick and how autonomous the work is, so two similar months can bill very differently, and teams migrating from another editor report spending several hours per developer adapting workflows and settings. None of this dents the raw capability — Cursor remains the most able tool in the category — but anyone on a fixed budget should track usage closely in the first month rather than assume the headline price is the bill.

Who is Cursor for?

Cursor is the right tool for professional developers and teams who want the most capable AI coding environment available and will use Composer and agents heavily enough to justify the cost — especially those working across large, real codebases in JavaScript, TypeScript and Python, where its whole-project understanding shines. If capability is the priority, nothing beats it. It’s the wrong fit if you want the cheapest serious option or fully predictable monthly billing — there, GitHub Copilot‘s flat entry price and unlimited free completions make more sense — or if you only need occasional autocomplete, in which case Cursor’s free Hobby tier is a fine no-cost starting point before you commit.

FAQ

AI codingCursorcode editorAI developer toolsAI tool review