Cursor: AI-Native IDE

Overview

Cursor is an AI-native IDE built as a fork of VS Code. It ships with AI features baked into the editor layer: inline completions (via the Supermaven autocomplete engine), a chat sidebar for codebase Q&A, and an agent mode that can read and write files, run terminal commands, and iterate on feedback.

It reached $2B ARR in early 2026 and is the highest-revenue AI coding product as of this writing.

Pricing

Tier	Price	Key limits
Hobby	Free	2,000 completions/mo, 50 slow premium requests
Pro	$20/mo	Unlimited completions, 500 fast premium requests/mo
Pro+	$60/mo	Unlimited fast premium requests
Ultra	$200/mo	20x the usage of Pro
Business	$40/user/mo	SSO, centralized billing, privacy mode

Premium requests use frontier models: GPT-5, Claude Sonnet/Opus, Gemini 2.5 Pro. "Fast" requests hit the frontier model directly; "slow" requests queue behind them.

Cursor 3.0: Agents Window

Cursor 3.0 introduced the Agents Window: parallel agent sessions running in the cloud, each in its own git worktree. Teams can run multiple agents on separate branches concurrently and review their diffs side by side before merging. Worktrees also enable agent sessions over SSH for remote development environments.

Autocomplete engine

Cursor uses Supermaven for autocomplete, which is a separate model optimized for low-latency next-token prediction. This is distinct from the chat and agent models; it runs on every keystroke rather than on demand.

Where it fits

Cursor is the IDE-native choice: you get the full VS Code ecosystem (extensions, keybindings, themes) plus tightly integrated AI at every layer. Teams that prefer a CLI-first workflow use Claude Code or Aider instead. Teams that want to stay in VS Code without switching editors use Cline or Continue as extensions.

The main trade-off against Claude Code: Cursor has the richer GUI experience and parallel agents, but it is a closed-source product. Claude Code is a terminal tool built on the open Anthropic API SDK.

Field notes

Builder.io engineering team benchmarked Cursor against Claude Code in May 2026 across five task types (refactoring, bug fixing, feature addition, test writing, documentation). Cursor led on tasks requiring frequent file navigation and diff review; Claude Code led on tasks well-suited to autonomous multi-step execution. [dev-post, 2026-05-01]