Image via Anysphere (Cursor) Official Site
Anysphere (Cursor)
Cursor 3
Released: 2026-04
Official Site →AI Tools
Cursor 3 Overview
Cursor 3 is an AI-native code editor built on a VS Code fork, developed by Anysphere and released in April 2026, targeting professional developers and engineering teams who want multi-agent workflows inside a familiar IDE environment. The tool supports parallel Background Agents that run in isolated VMs, open pull requests autonomously, and can be triggered remotely from Slack or GitHub while the developer's machine is offline. Its multi-model architecture, spanning Claude Sonnet 4.6, GPT-5.4, Gemini 3.1 Pro, and Anysphere's own Composer 2, separates it from single-provider AI coding assistants.
Pros
- Background Agents run in isolated VMs on separate Git branches and can open pull requests autonomously, removing the need to stay inside the editor while long tasks complete.
- Cloud Agents can be triggered from Slack, GitHub, or mobile and continue running with the laptop closed, which extends async workflows beyond the desktop.
- The Pro plan launches at $20/month and includes a matching $20 API credit pool plus unlimited Tab completions, giving solo developers a predictable cost floor.
- Students get one full year of Cursor Pro at no charge with a verified school email, covering the entire Pro feature set.
- Design Mode lets developers annotate UI elements directly in the browser, giving agents a concrete visual target rather than a text description alone.
- The multi-model setup spans Claude Sonnet 4.6, GPT-5.4, Gemini 3.1 Pro, and the in-house Composer 2, so users are not locked to a single provider's capabilities or pricing.
Cons
- Web search is not a native feature, putting Cursor 3 behind AI coding tools that can pull live documentation or check for recent API changes without a workaround.
- SOC 2 and GDPR compliance status are unconfirmed from available sources, which may block adoption in regulated industries where those certifications are required.
- The credit system means Pro users who exceed their $20 monthly pool face pay-as-you-go overage charges at full API token rates, and output tokens run at ~$6.00/M tokens, so heavy agent usage can escalate costs quickly.
- Org-wide privacy mode is gated behind the Teams plan at $40/user/month; Pro users can enable privacy mode individually but cannot enforce it across a team.
- The editor is a VS Code fork, which means JetBrains users (IntelliJ, PyCharm, WebStorm) have no native integration path and must switch editors entirely.
- The Ultra tier is priced at $200/month per user, a significant cost for individual developers whose workloads may not reliably justify the 20x credit multiplier over the $20 Pro tier.
Specifications
| Tool Category | Coding / AI-native IDE (multi-agent) |
| Version | 3 (major release, April 2, 2026) |
| Base Editor | Built on VS Code (fork) |
| Model Name(s) | Multi-model — Claude Sonnet 4.6, GPT-5.4, Gemini 3.1 Pro, Cursor Composer 2 (in-house, launched March 19, 2026) |
| Auto Mode Pricing | ~$0.25/M tokens (cache read), ~$1.25/M tokens (input), ~$6.00/M tokens (output) |
| Reasoning Mode | Yes — integrated via Composer 2 and frontier model selection |
| Code Sandbox | Yes — Background Agents run in isolated VMs on separate Git branches |
| Web Search | No (not a native feature) |
| Multimodal Inputs | Text, code, file upload, codebase context, UI annotation via Design Mode |
| Output Types | Code, code explanations, pull requests, multi-file refactors |
| Agents Window | Dedicated full-screen workspace for running and managing multiple AI agents in parallel |
| Background Agents | Yes — run in isolated VMs, open pull requests autonomously when finished |
| Cloud Agents | Yes — triggered from Slack, GitHub, or mobile; continue running with laptop closed |
| Design Mode | Yes — annotate UI elements directly in the browser to give agents precise targets |
| M C P Support | Yes (Pro and above) |
| Skills And Hooks | Yes (Pro and above) |
| S S H Support | Yes — agents run across local machines, cloud environments, and SSH connections |
| A P I Access | Yes — usage-based, billed at API token rates per model |
| Free Tier | Yes — Hobby plan, no credit card required; limited Agent requests and limited Tab completions |
| Paid Plan — Pro | $20/month; includes $20 API usage credit pool, unlimited Tab completions (Auto mode), frontier model access, cloud agents, MCPs, skills, hooks |
| Paid Plan — Pro+ | $60/month; everything in Pro plus 3x usage credits on all OpenAI, Claude, Gemini models |
| Paid Plan — Ultra | $200/month; everything in Pro plus 20x usage credits and priority access to new features |
| Paid Plan — Teams | $40/user/month; Pro-equivalent AI access plus shared chats, commands and rules, centralized billing, usage analytics and reporting, org-wide privacy mode controls, role-based access control, SAML/OIDC SSO |
| Enterprise | Yes — pooled usage across org, custom pricing via sales |
| Annual Billing Discount | 20% across all paid plans (effective ~$16/month for Pro) |
| Student Access | Yes — one year of Cursor Pro free with verified school email |
| Platforms | Windows, macOS, Linux (VS Code-based); mobile agent launch support |
| V S Code Extension Compatible | Yes (built on VS Code fork) |
| Privacy Mode | Yes — when enabled, code data is never stored by model providers or used for training |
| Privacy Mode Scope | Available org-wide on Teams plan |
| Overage Billing | Yes — pay-as-you-go at same API rates after credit pool exhausted |
| Credit System | Usage-based since June 2025; each plan includes a monthly credit pool equal to plan price in dollars |
Specifications should be verified from official manufacturer sources. Details may vary by region or configuration.
Disclaimer: Specifications, pricing, and availability are subject to change.
Please verify all information from the official manufacturer website before purchasing.