Cursor's free tier is called Hobby. It needs no credit card, does not expire, and gives you Agent, Chat, and Tab completions on the Auto model with limited usage. Cursor does not publish a fixed request count for it, so any article quoting one is describing a retired pricing era. Pro at $20 a month adds unlimited Tab, extended agent limits, a $20 allowance for third-party models such as Claude and GPT, and cloud agents.
- Hobby is a real tier, not a timed trial. No credit card, no expiry.
- You get Agent, Chat, and Tab on the Auto model, with limited usage.
- Cursor publishes no fixed number. Read the usage view in your own account.
- The agent allowance bites long before the completion allowance does.
- Upgrade when you hit a limit mid-task, twice. Not on principle.
Hobby against Pro
From the Cursor pricing page, the pricing reference, and the pricing help article, checked August 2026.
| Hobby (free) | Pro ($20/mo) | |
|---|---|---|
| Credit card | Not required | Required |
| Tab completions | Limited | Unlimited |
| Agent requests | Limited | Extended limits |
| Model choice | Auto | Auto, Composer, Grok, and frontier models |
| Third-party allowance (Claude, GPT, Gemini) | Not included | $20 a month |
| On-demand usage past the allowance | No | Yes, at model API rates |
| Cloud agents | No | Yes |
| MCP servers, skills, hooks | No | Yes |
| Bugbot | No | Yes, on usage-based billing |
| Time limit | None | Not applicable |
Why nobody can tell you the exact limit
Search this question and you will find confident numbers: two thousand completions, fifty slow premium requests, five hundred fast requests. Those numbers were real once. They belong to a pricing model Cursor retired.
The eras, and what each one measured.
| Era | Unit | Why the numbers stopped meaning anything |
|---|---|---|
| Fast and slow requests | A monthly count of premium requests | Retired. The complaint was running out mid-month and queueing behind slow requests |
| Credit pool | Compute credits by model | Retired. Opaque, and hard to predict before you spent them |
| Current | Two pools: included Cursor Models, plus a dollar allowance for other models | Hobby has no dollar allowance at all, so there is no figure to publish |
Cursor's current public pages describe Hobby as core features with limited usage: Agent, Chat, and Tab completions on the Auto model, with no numeric quota attached. That is less evasive than it looks. Usage on paid plans is measured in tokens against a dollar value, and Hobby sits outside that system entirely, so there is no dollar figure to quote.
Read your own number instead
Open Cursor and check the usage view in the app, or sign in to the Cursor dashboard on the web. That shows the allowance attached to your account, which is the only number that governs your day.
Watch which pool empties first
On any plan the agent allowance runs out long before the completion allowance does, because an agent turn reads files, runs commands, and re-reads its own output. A completion is one short request against a model built to be cheap.
Distrust any article that gives a number without a date
Including this one, if it ages. Cursor has changed the shape of this three times, and a figure written into a page in 2025 is not a figure, it is an artefact.
What you can genuinely do on Hobby
- Evaluate Tab properly. Completions are the cheapest thing Cursor does and the free allowance goes furthest here. Tab is the feature worth evaluating, so this is the good news.
- Learn the editor. Keybindings, Inline Edit, Plan mode, the rules system, the command palette. None of that is gated behind a plan.
- Run a side project. Light and intermittent use fits comfortably inside the free allowance.
- Decide about the agent. You will hit the agent limit quickly, and that is itself the answer to whether you need Pro.
What Hobby does not cover is daily professional use with the agent. That is deliberate, and it is the right constraint: agent turns are what actually cost Cursor money, and a free tier that funded them would not exist. Hobby also runs on Auto rather than letting you pin Claude or GPT, so third-party frontier models are a paid feature.
Making the free tier last
- Stay on Auto. It is what Hobby gives you anyway, and on paid plans it is the setting that keeps the metered pool untouched.
- Use Plan mode before Agent. Shift and Tab from the chat input. Reading a plan costs almost nothing and stops the agent spending a dozen turns going the wrong way.
- Scope the request. Name the files. An unscoped request makes the agent read half the repository before it edits anything, and reading is what consumes the turn.
- Write one rules file. Conventions stated once in
.cursor/rules/orAGENTS.mdcost far less than the same corrections issued in every prompt. - Do the mechanical edits with Tab. Renaming, propagating a signature change, filling in a switch statement: Tab is better at these than the agent, and cheaper.
When to upgrade, and when not to
| Signal | Verdict |
|---|---|
| You hit a limit mid-task, twice | Upgrade. Once is bad luck, twice is a pattern, and lost flow is worth more than $20 |
| You avoid the agent to conserve requests | Upgrade. Rationing a tool you are evaluating means you are no longer evaluating it |
| You have used it daily for a month | Upgrade. It is part of your workflow and the price is no longer the question |
| You want Claude or GPT specifically | Upgrade. Hobby runs on Auto; the third-party allowance starts at Pro |
| Two months on Hobby, never hit a limit | Do not upgrade. You are a completion user and Pro sells allowance you will not spend |
| You are exhausting Pro every month | Compare the current unpublished Pro Plus and Ultra prices in Cursor's plan picker against what your overage costs |
Free alternatives worth knowing
Checked August 2026.
| Tool | Free tier | Good for |
|---|---|---|
| Cursor Hobby | Limited agent, Auto model, no expiry, no card | Evaluating Tab and the editor |
| GitHub Copilot | Limited free plan; free for verified students | Completion in the editor you already use |
| Gemini CLI | 60 requests a minute, 1,000 a day on a personal Google account | Real terminal agent work at zero cost |
| Aider | Free software, you pay tokens | Control, no subscription, pennies per session |
| Claude Code / Codex | No free tier | Entry at $20 either way |
Questions people ask
Yes, there is a free tier called Hobby. It needs no credit card, does not expire, and gives you Agent, Chat, and Tab completions on the Auto model with limited usage. Pro is $20 a month and raises the limits substantially.
Limited Tab completions and limited agent requests, on the Auto model, with no dollar allowance for third-party models such as Claude, GPT, and Gemini. Cursor does not publish a fixed number, so check the usage view in your own account rather than trusting an article.
No. Hobby is a tier rather than a trial, which is exactly why it is useful for deciding whether the editing model suits you over a month of real work.
Because Cursor has changed its pricing shape three times: fast and slow request counts, then a credit pool, then the current two-pool arrangement. Figures like "2,000 completions and 50 slow requests" belong to a retired era.
No. Hobby runs on Auto. The third-party model pool starts on Pro, which includes a $20 monthly allowance for it, and continues on demand at API rates past that.
When you hit a limit mid-task twice, or when you notice yourself avoiding the agent to conserve requests. If you have gone two months without hitting a limit, do not upgrade.
Cursor does not publish a standing Pro trial on its pricing pages as of August 2026. If you are offered one at sign-up, spend it on agent work rather than on Tab, since Tab is already well covered by the free tier.
Not any more. Cursor discontinued new sign-ups for its student discount on 25 June 2026. Undergraduates can get credits at Cursor events, and graduate students, researchers, and educators can request credits through a form.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.