Cursor vs. GitHub Copilot

Editor's note: This is a documentation-led edition. We have not yet run our own hands-on tests of Cursor or GitHub Copilot. What follows rests on official documentation, published pricing and independent studies we read in full, all accessed on 16 October 2026. A tested update will follow.

Cursor and GitHub Copilot now sell almost the same features: completion, an editor agent, cloud agents that open pull requests, and an AI reviewer. The differences are in where each lives, what it does before asking, what happens to your code and how you pay.

This comparison sets those points side by side from each vendor's own pages. Our shortlist of AI coding assistants covers five more.

What each is for

Cursor is an AI-first code editor. Its documentation says it "is based upon the VS Code codebase", so VS Code settings and extensions import in one click.[1] Around it sit a CLI, cloud agents you can start from the web, Slack, GitHub or an iPhone app, and Bugbot, a pull-request reviewer.[2][3]

GitHub Copilot is an assistant that plugs into other people's editors: VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Neovim, Zed and others. It also lives on GitHub.com, in GitHub Mobile and in the Copilot CLI.[4] Its cloud agent and code review run inside GitHub's pull-request workflow.[5][6]

So the first question is whether your team will switch editors.

Side by side: documented facts (16 October 2026)

Sources: Cursor pricing and models pages; GitHub plans page and docs.[7][8][4][9] US list prices, excluding tax.

The editor model: switch editors, or keep yours

Cursor asks you to move into its editor. Because it shares VS Code's base, the move is short for VS Code users and longer for anyone on JetBrains, Xcode or Visual Studio.[1]

Copilot asks nothing of the kind: GitHub documents agent mode in VS Code, Visual Studio, JetBrains IDEs, Xcode and Eclipse, and completion in more editors still.[4][10] The catch: features differ by editor. Content exclusion, for instance, is "not supported in Edit and Agent modes" of Copilot Chat in VS Code and other editors.[11]

The agent in your editor

Both agents take a task, choose files, edit them and run commands. Their defaults differ.

Cursor. The agent "can modify workspace files without approval, except for configuration files", and changes save straight to disk.[12] On terminal commands, Cursor's pages disagree. Its agent-security page says "by default, terminal commands need your approval".[12] Its enterprise controls page says that "in Cursor 3.6 and above", Auto-review is the default Run Mode.[13] In Auto-review, allowlisted commands run, others run in a sandbox where possible, and a classifier model on Cursor's servers judges the rest. Cursor's own heading is blunt: "Auto-review is not a security boundary".[14] Check Settings › Agents › Approvals & Execution.

Copilot in VS Code. GitHub says that "by default you approve commands before they run".[10] VS Code's own documentation is more precise: "common read-only commands run automatically, while risky commands such as rm and del require approval". The default permission level, Manual, asks for anything not auto-approved. "Allow all" and the Autopilot mode skip prompts, and organisations can hide those options.[15]

In both, an allowlist or classifier is a convenience, not a wall.

Cloud agents: work that happens while you are away

Both products can take a task, work on it in a remote machine and hand back a branch or pull request.

Cursor's cloud agents "run in isolated VMs in the cloud" on Cursor's AWS infrastructure, clone your repository from GitHub, GitLab, Bitbucket or Azure DevOps, and push to a separate branch.[2][16] Two defaults matter. "The agent has internet access by default", which admins can restrict by domain. And unlike the editor agent, it "auto-runs all terminal commands", which Cursor itself says "introduces data exfiltration risk".[16] Conversation history is kept indefinitely by default; environment snapshots are deleted after 30 days unused.[16]

Copilot cloud agent edits files and runs tests "in an ephemeral cloud development environment" on GitHub.[5] Its internet access is limited by a firewall by default, with a recommended allowlist for package registries. GitHub warns that the firewall can be bypassed and covers only processes the agent starts.[17]

The bigger difference is who may merge. GitHub limits the cloud agent to a single branch and plain pushes, and holds GitHub Actions workflows until a person approves them. The person who asked for the change cannot approve it.[18]

"Copilot cloud agent cannot mark its pull requests as "Ready for review" and cannot approve or merge a pull request."

— GitHub Docs, Risks and mitigations for GitHub Copilot cloud agent (2026).[18]

Cursor's cloud agents sign every commit and follow whatever branch protection you set.[2][16] Copilot inherits extra guardrails from GitHub; with Cursor, you build them.

Code review: Bugbot vs. Copilot code review

Bugbot reads pull-request diffs and leaves comments with fixes, on every update or when someone comments "bugbot run". It works with GitHub, GitLab, Bitbucket, Azure DevOps and Cursor's own Origin hosting. Its findings post as a "neutral" check by default, so requiring the check does not block a merge unless you turn on failing for unresolved issues.[3]

Copilot code review runs on GitHub.com, in IDEs and, in preview, on Azure DevOps. It uses GitHub Actions to gather wider repository context. GitHub estimates a review at $0.05 to $1 in credits on "Lite" effort and $0.25 to $5 on "Balanced", plus Actions minutes since 1 June 2026. You cannot pick its model.[6] By default, its reviews "do not count toward required approvals"; a preview setting lets admins allow Copilot approvals.[6]

Neither reviewer, at default settings, replaces a human approval.

Model access

Cursor sells its own models inside the plan. The usage pool covers "Grok 4.7, Grok 4.6, Grok 4.5, and Composer 2.5". Plus is limited to these. On Premium and above, third-party models such as Claude, GPT and Gemini are available "at the model's API list price", paid from prepaid credit.[8] Even with your own API key, Cursor says requests "still go through our backend".[19]

Copilot lists models from Anthropic, OpenAI, Google, xAI and Microsoft, plus Kimi models, on paid plans, drawn from the same credit allowance at different rates. Free and Student plans get automatic model selection only. Pro+ and Max can also hand tasks to Anthropic's Claude and OpenAI Codex agents, in preview.[9][4]

In short: Copilot's third-party models come out of plan credits; Cursor bills them on top of a pool that covers Grok and Composer.

Plans and prices

Cursor: now part of The X Subscription

We re-checked this on 16 October 2026. Cursor's models page says "Cursor is sold as part of The X Subscription, one subscription for Grok, Cursor, and X."[8] The pricing page lists X features (a premium checkmark, fewer ads, X Money cashback) beside Cursor features. Its FAQ says: "Subscriptions are only sold directly through X Subscription Corp."[7]

Prices as of 16 October 2026: Free; Plus $8 a month (Grok models only, no cloud agents or Bugbot); Premium $30; Super $100; Ultra $200. Annual billing is $80, $300, $1,000 and $2,000 a year. Teams seats are $30, $100 or $200, and third-party model use on Teams adds a "Cursor Token Rate". Enterprise is custom.[7][8] The pool "refreshes weekly" and is shared across Cursor and Grok.[7] Older Cursor plans (Pro, Pro+, Ultra, Teams Standard and Premium) are labelled "legacy" and keep working.[8]

For procurement, that is the headline: the contracting party is X Subscription Corp, and the subscription includes social-media features.

Copilot: monthly AI credits

Prices as of 16 October 2026: Free; Pro $10 a month; Pro+ $39; Max $100; Business $19 and Enterprise $39 per seat.[4][9] Chat, agents, code review and the CLI use GitHub AI Credits at "1 AI credit = $0.01 USD". Completions and next-edit suggestions do not use credits on paid plans.[4] Paid individual plans add a "flex allotment": Pro carries 1,500 credits in total, Pro+ 7,000 and Max 20,000. Business includes 1,900 credits per user and Enterprise 3,900.[9]

Data use and training

GitHub's change. On 25 March 2026, GitHub announced that from 24 April, interaction data from Copilot Free, Pro and Pro+ would train its models unless users opt out.[20] Its documentation now lists Max as well.[21] The data includes inputs, outputs, code snippets, file names, repository structure and feedback. It may be shared with affiliates "including Microsoft", but not with third-party model providers. Earlier opt-outs carry over. Private repositories are not used "at rest", but code you work on with Copilot counts:

"This interaction data is required to run the service and could be used for model training unless you opt out."

— GitHub Blog, Updates to GitHub Copilot interaction data usage policy (2026).[20]

The opt-out is "Allow GitHub to use my data for AI model training" in Copilot settings.[21] Business and Enterprise data is not used for training. For those plans, IDE chat and completion prompts are not retained; other surfaces keep them for 28 days.[4]

Cursor's Privacy Mode. With it on, "Customer Data will not be used for training by Cursor", and model providers work under zero-data-retention agreements. With it off, Cursor "may use and store codebase data, prompts, editor actions, code snippets" to train its models.[19] It is on by default for Enterprise teams, and team admins can enforce it.[22] The pages we read do not say whether it is on by default for individuals. A few models need provider retention and require admin approval under Privacy Mode.[22]

Enterprise controls

Copilot Business and Enterprise. Admins set policies for features, agents and models at enterprise or organisation level, and the audit log records policy changes.[23] Content exclusion stops Copilot reading named files, with the editor gap noted above.[11] A public-code filter blocks suggestions that match public code, and GitHub's FAQ ties IP indemnity to unmodified suggestions "when Copilot's filtering is enabled".[4] Microsoft's copyright-commitment page, however, says that from 3 April 2026 the duplicate-detection filter "is no longer required for CCC coverage" of GitHub offerings.[24] Until the two agree, check your contract or ask your account team which applies, and keep the filter on. Copilot is not available for GitHub Enterprise Server.[9]

Cursor Teams and Enterprise. Teams adds central billing, usage analytics, team-wide privacy settings and SAML/OIDC single sign-on. Enterprise adds SCIM, audit logs, spend controls and admin control of auto-run, browser, network, repository, model and MCP access.[7][25] Hooks can block prompts, file reads or commands; customer-managed encryption keys and US-only data residency are available.[13][22] Cursor lists SOC 2 Type II, ISO/IEC 27001 and ISO/IEC 42001.[26]

What independent research says

No independent study has compared current versions of the two. METR's 2025 randomised trial found experienced developers took 19% longer with AI tools, "primarily Cursor Pro", and METR now labels the results "out of date".[27] Field experiments by Cui and colleagues found a 26.08% rise in completed tasks (standard error 10.3%) with 2022–23 Copilot autocomplete.[28] Our workplace AI evidence review covers both.

Benchmarks do not settle it either. In February 2026 OpenAI stopped reporting SWE-bench Verified, the coding score most launch posts quoted, after finding flawed tests and memorised answers.[29] Our benchmarks explainer explains why one score cannot rank a tool.

Documented limits and risks

• Cursor's terminal default is unclear across its own pages: approval in one, Auto-review in another.[12][13]
• Cursor cloud agents run every command and reach the internet unless you restrict them.[16]
• Copilot content exclusion has gaps: not in agent and edit modes, not for symlinks.[11]
• Individual plans train by default on Copilot; Cursor may train whenever Privacy Mode is off.[20][19]
• Costs are usage-based on both, so heavy agent use can exceed the headline price.[4][8]

Who it suits, and who should skip it

Cursor suits developers who want an agent-first editor, are happy in a VS Code-style environment, and host code on GitHub, GitLab, Bitbucket or Azure DevOps. Skip it if the X Subscription contract will not pass procurement, if you need Claude or GPT inside a flat price, or if your team lives in JetBrains or Xcode.

Copilot suits teams whose code and reviews already live on GitHub, and developers who want AI in the editor they already use. Skip the individual plans for client or employer code unless training is switched off, and skip Copilot if you run GitHub Enterprise Server.

For a deeper look at each, see our Cursor review and GitHub Copilot review.

AI Chronicle

How to test it yourself

1. Use one repository and one task list for both: a failing test, a small feature, a refactor under tests.
2. Record the defaults first: Cursor's Run Mode and Privacy Mode; Copilot's permission level and training setting.
3. Work on a branch with no production secrets, and turn on Cursor's workspace trust for unfamiliar code.
4. Count prompts and rejections: how often each asked, and how often you said no.
5. Try one cloud-agent task each, with internet limited to your package registry.
6. Run both reviewers on the same pull request and log which comments were right, wrong or noise.
7. Time the whole job, review and rework included, as our guide to measuring AI productivity describes.
8. Track cost per task in Cursor's usage view and Copilot's credit meter.

A worked example (hypothetical)

This scenario is invented to show how the documented differences play out. No tool was run, and no outputs are reported.

Priya runs engineering at a 12-person agency. Code lives on GitHub; half the team uses VS Code, half JetBrains.

Data first. Client contracts forbid training on their code, so personal Copilot Pro or Cursor without enforced Privacy Mode are out. Both Copilot Business and Cursor Teams with enforced Privacy Mode pass.

Editors next. Cursor would move the JetBrains half into a new editor. Copilot Business works in both.

Merge control. Priya wants agents to open pull requests but never merge. Copilot's rules do that by default; with Cursor she would set branch protection herself.

Procurement. For Cursor, the contracting party is X Subscription Corp.

So she pilots Copilot Business first, with a Cursor pilot for the VS Code half. The choice rests on terms; speed is still unmeasured.

Questions readers commonly ask

Which is better, Cursor or GitHub Copilot?

We have not tested them, so we do not pick a winner on output. On documented terms, Copilot fits teams already on GitHub; Cursor fits developers who want an agent-first editor.

Do I need an X account to use Cursor?

Cursor's pages say it is sold as part of The X Subscription through X Subscription Corp.[7][8] They do not say plainly whether an X account is required. Check at sign-up.

Can I stop GitHub training on my code?

Yes, on individual plans: set "Allow GitHub to use my data for AI model training" to Disabled in your Copilot settings. Business and Enterprise data is not used for training.[21][4]

Is either safe to let run commands unattended?

Only within limits you set; both vendors call their allowlists and classifiers imperfect. Our research on what agent platforms disclose about permissions compares the disclosures.

The next step

Considering a terminal agent instead? Read our Claude Code vs. Cursor comparison next.

Sources and evidence note

Product descriptions, defaults, data terms and prices come from Cursor's and GitHub's own documentation, pricing, blog and security pages, and from Microsoft's VS Code documentation. They describe intended behaviour, not tested performance. Prices are US list prices as displayed on 16 October 2026. Independent evidence comes from METR, Cui and colleagues, and OpenAI's benchmark post, each read on the publisher's site. The worked example is invented. All pages were accessed on 16 October 2026.

Related reading

• Claude Code review: from task to reviewed change
• What workplace AI studies actually show