Cursor 3.0

Does not work from mainland China

Cursor restricts all U.S.-hosted AI models (Claude, GPT-4, Gemini) in this region due to compliance requirements. Students on campus or using a Chinese network will see “This model provider doesn’t serve your region” and cannot use Cursor.

If you are studying from campus: use GitHub Copilot instead — it works reliably from mainland China via the VS Code extension.

If you are off-campus with reliable access to international networks: Cursor is worth setting up.

---

What is Cursor 3.0?

Cursor 3.0 (released April 2, 2026) is a major architectural update to the Cursor IDE — a VS Code fork built around AI agents. It shifts from a single chat interface to a full multi-agent workspace: you can run several agents in parallel, each working in an isolated environment.

This makes it qualitatively different from GitHub Copilot. Copilot assists you. Cursor 3.0 agents work independently on tasks while you review and direct them.

---

What is new in Cursor 3.0

Agents window — parallel agents

You can now run multiple agents at the same time in separate isolated workspaces. Each agent can work on a different task — one refactoring a component, another writing tests, another debugging an error — without interfering with each other.

For a student team, this means different team members can direct separate agents on separate features simultaneously.

Design Mode

Point at any UI element in the browser. The agent sees exactly what you see and makes precise edits to the code that produced it. This is the fastest way to iterate on a frontend component — show it the problem, describe the fix.

/worktree and /best-of-n

Command	What it does
/worktree	Runs the agent in an isolated git branch — changes are safe to discard if the result is wrong
/best-of-n	Runs the same task on multiple models in parallel and shows you the results side by side — you pick the best one

/best-of-n is particularly useful when you are not sure how to solve a problem: give it to three models, compare their approaches, choose what you understand best.

BugBot

BugBot reviews your pull requests automatically, identifies bugs, and can open a cloud agent to propose fixes. If your team uses pull requests on GitHub, BugBot runs in the background on every PR.

---

Pricing

Cursor has a free tier with a monthly cap on fast requests. Beyond that cap, it falls back to slower models.

Bringing your own API key is a Pro feature. On the free tier, you cannot swap in China-accessible models (Qwen, DeepSeek, etc.) — you are limited to Cursor’s hosted models, which are U.S.-based and blocked in mainland China. The free tier does not solve the China access problem.

Student Pro — not available for Chinese university emails

Cursor offers one free year of Pro with a verified university email. XJTLU email addresses (.edu.cn) are not accepted — the student verification rejects Chinese university domains. This option is not available to most students in this course.

If you have a non-Chinese university email from an exchange program or dual degree: it may work — check cursor.com/students.

---

Cursor vs GitHub Copilot — when to use which

Situation	Use
You are on the XJTLU campus network	Copilot — Cursor does not work
You need inline completions while typing	Copilot — Cursor’s completions exist but Copilot’s are smoother
You want to build an entire feature from a description	Cursor — its agents handle more files and more complexity
You are iterating on a UI and want to point at the problem	Cursor Design Mode — nothing else does this
You want to try two approaches and pick the better one	Cursor /best-of-n — runs multiple models in parallel
You are not sure what is wrong with your code	Either — Cursor is stronger for multi-file reasoning; Copilot is sufficient for single-file debugging

---

Setup

1. Download from cursor.com — available for macOS, Windows, Linux
2. Open Cursor → it imports your VS Code extensions and settings automatically
3. Sign in with your student account
4. Start with Cmd/Ctrl + I to open the Composer — describe a task in plain English

Start with a specific, bounded task

Like all AI agents, Cursor works best on tasks with a clear input and expected output. “Build my whole app” will produce something you cannot understand or maintain. “Add a character counter below this text input that shows X/200” will produce something you can verify, explain, and ship.