Claude Code vs Cursor: Claude Now Costs Extra in Cursor
On this page, 7 sections
Pick Claude Code if you live in the terminal and want Claude on a flat plan. Pick Cursor if you want an editor and you are happy with Grok models in the price. What changed on 12 October 2026 is that Claude inside Cursor is no longer covered by the Cursor subscription. It is billed at API list price from the first request.
I use Claude Code every day and I tried Cursor only briefly, so this is not a speed or code quality test. It is a read of the docs, the prices and my own setup. Every source below was opened on 18 November 2026.

What changed in Cursor on 12 October
Cursor now belongs to SpaceX. The New Stack reports the deal closed in mid-August 2026, and on 12 October the company announced The X Subscription, one plan that bundles Grok, Grok Bot, Cursor and X Premium. Cursor’s own pricing page now sells Cursor through that plan.
The tiers, from the announcement and from Cursor’s docs: Free at $0, Plus at $8 a month, Premium at $30, Super at $100 and Ultra at $200. Every paid tier includes a weekly pool of usage for Grok models and Cursor’s own Composer model, shared across Grok and Cursor.
Here is the part that matters if you picked Cursor for Claude. Cursor’s docs say third-party models like Claude, GPT and Gemini are not part of the included pool. On Premium and above they run at the model’s API list price from the first request, paid from prepaid credit. On Free and Plus they are not available at all.

The New Stack worked out what that means. Before, a $20 Cursor Pro plan included $20 of usage on Claude, GPT and Gemini at API prices, so a developer who used $20 of Claude paid $20 in total.
Under the new Premium plan the same usage is $30 for the subscription plus $20 of model charges, so $50. The same article notes that cloud agents and Origin, Cursor’s GitHub alternative, need Premium too, so someone coming from the old Pro plan pays $10 more for those.
Two things soften this, and both are in the docs. Existing Cursor plans, which the docs call legacy plans, keep working as they are unless you switch.
And SpaceX says each X Subscription plan includes at least twice as much coding usage with its own models as the comparable Cursor plan. That is the company’s claim, not a measurement, and it only helps if you are happy to work with Grok models.
What Claude Code costs by comparison
Claude Code comes with a Claude subscription. Anthropic’s pricing page lists it on Pro, which is $20 a month billed monthly or $17 a month on an annual plan, and on Max at $100 a month with 5x the Pro usage.
There is also an Ultra tier at $200 a month. Every plan has usage limits that reset on a five hour window with weekly limits on top, so the price is flat but the allowance is not unlimited.
That is the real difference in how the bill behaves. With Claude Code you pay a flat plan and hit a limit. With Claude in Cursor you now pay a flat plan for the editor and then pay per use for Claude on top.
| Setup | What you pay for Claude | How it behaves |
|---|---|---|
| Old Cursor Pro, $20 of Claude usage | $20 in total | Usage was inside the plan |
| Cursor Premium, $20 of Claude usage | $30 plan plus $20 usage, $50 | Each request billed at API price from prepaid credit |
| Claude Pro with Claude Code | $20 a month, or $17 on annual billing | Flat price, usage limits apply |
Rules, hooks and the part I care about most
Price is the headline, but my own choice has never come from price. I care about whether the rules I write down are followed. I wrote about that for Codex in Claude Code vs Codex, and the same test works here.
Cursor’s rules live in .cursor/rules as .mdc files with a header that says whether a rule always applies, applies to certain file paths, or is picked up when its description matches. A plain .md file in that folder is ignored. Cursor also reads AGENTS.md in the project root and in subfolders as a simpler option.
Claude Code reads CLAUDE.md, a global file for every project and a project file for each repo. The shape is the same: write your working agreements once in markdown and keep them short. My own CLAUDE.md is a handful of behavioural rules. Anything that must not be forgotten lives in a hook, which I explained in Claude Code hooks explained.
Both tools now have hooks, and this is where Cursor surprised me. Its docs list sessionStart, preToolUse, postToolUse, beforeShellExecution, beforeMCPExecution, subagentStart and stop, with a rule that a permission hook can deny a call. Claude Code’s hook docs cover SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, PermissionRequest, Stop and more. So the old argument that only the terminal tool can enforce rules is out of date.
One limit in the Cursor docs is worth knowing. For cloud agents, a sessionStart hook is deferred, because the agent can start in a read-only environment before hooks load. If your safety rule must run before the first write, check that case before you trust it. I did not run it.
| What I check | Claude Code | Cursor |
|---|---|---|
| Where you work | Terminal, no editor | An editor built on VS Code, plus a CLI and cloud agents |
| Models | Claude models | Grok models in the plan, Claude and others at API price |
| Instruction files | CLAUDE.md | .cursor/rules and AGENTS.md |
| Hook that can block a tool call | PreToolUse | preToolUse, beforeShellExecution |
| Billing shape | Flat plan with limits | Flat plan plus per use for third-party models |
How I would decide today
- You want Claude and a predictable bill. Claude Code on a Claude plan. You will hit a limit before you hit a surprise invoice.
- You live in an editor and review every diff. Cursor. roadmap.sh describes the split as Claude Code being the one you hand work to and Cursor being the one you stay with, and that matches how I think about it.
- You mostly use Grok models already. The new Cursor plans are built for you. The included pool covers them and it is the one place the bundle is clearly good value.
- You switch models by task. Cursor still lets you, but every Claude or GPT request is now metered. Work out your monthly usage before you move.
- You run agents in scripts or on a server. A terminal tool fits better. Cursor has a headless CLI too, but I have not tried it, so I will not rank it.
My view, and it is an opinion rather than a test: Cursor earned its place by being the one editor where you could use the best model from anyone. A plan that puts one vendor’s models inside the price and meters the rest changes what it is.
A Cursor user quoted by The New Stack says much the same, that a lot of developers keep Cursor, Codex and Claude Code all installed and keep Cursor open all day because it is not tied to one vendor. Anthropic has said it will keep supporting Claude in Cursor, per the same report, so the access is not going away. The price of it is.
If you are building a setup around agents, the editor matters less than the layer under it. The same instruction file, the same hooks and the same MCP servers should work wherever you sit. I covered that side in what an MCP server is and in MCP vs CLI for AI agents.
What I could not verify
- Which tool writes better code. I did not benchmark them and I will not put a number on it.
- How much of the included Grok pool a normal week uses. The docs say it refreshes weekly and different models draw at different rates, and I have not run a week on it.
- Whether my own hooks run unchanged inside Cursor. I read its hook docs and did not port my scripts.
- Prices move. These are the numbers on 18 November 2026, and both vendors say prices can change.
FAQ
Can I still use Claude models in Cursor?
Yes, on the Premium, Super and Ultra tiers. Cursor’s docs say third-party models such as Claude are available at the model’s API list price, paid from prepaid credit. They are not available on the Free and Plus tiers.
Do old Cursor plans change?
Not automatically. The docs call Pro, Pro+, Ultra, Start and the older Teams seats legacy plans that keep working as they are. The X Subscription docs say switching is optional and that you get a prorated refund if you do switch.
Is Claude Code cheaper than Cursor if I mostly use Claude?
For a light user the flat Claude plan is simpler and often cheaper, because Cursor now adds per use Claude charges on top of its own fee. The two are not measured in the same units, though, so a heavy user should check their own usage on both sides before deciding. I have not run that comparison on my own bill.
Sources
- Introducing The X Subscription, x.ai, 12 October 2026
- The X Subscription pricing docs
- The New Stack, The X Subscription bundles Cursor with Grok and X Premium. But Claude now costs extra, 13 October 2026
- Cursor docs, models and pricing
- Cursor docs, rules and hooks
- Claude pricing, Anthropic
- Claude Code docs, hooks
- roadmap.sh, Claude Code vs Cursor