What it is
Cursor is a code editor forked from VS Code with AI built into every part of it: tab completion, chat, and agents that plan and edit across many files. In 2026 the app has been rearranged around an Agents Window, where you run several agents side by side, open subagents as tabs, and hand work to cloud agents that run in their own virtual machines.
SpaceX completed its purchase of Cursor's maker, Anysphere, on 14 August 2026, and Cursor is now sold as part of the X Subscription. One weekly usage pool covers Cursor's own models (Grok 4.7, 4.6, 4.5 and Composer 2.5) and the Grok apps. Claude, GPT and Gemini sit outside the pool and are billed at API list price.
Around the editor sit Bugbot (automatic pull request review), a Security Reviewer agent, Automations, a CLI, and Origin, a code hosting and pull request service.
Our review
Handing a coding job to an AI usually means babysitting it: one task at a time, in one window, while you wait to see what it broke.
Cursor lets you hand out several jobs at once. Its Agents Window runs agents side by side, and cloud agents work in their own virtual machines, so they keep going after you close the laptop. CNBC reported in February 2026 that those agents can test their own changes and record video, logs and screenshots of what they did, which means you check a recording instead of rerunning everything yourself.
It also moves faster than any rival editor. In our reading of the release notes, almost every week adds something. October 2026 alone brought live progress for subagents, a stricter check before an agent deletes anything risky, and admin controls over which MCP tools (plug-ins that let agents reach other apps) your team may use.
The catch: who owns Cursor now decides what you pay. SpaceX closed its $60 billion all-stock purchase of Anysphere, Cursor's maker, on 14 August 2026, and you now buy Cursor through X Subscription Corp, bundled with Grok Chat, Imagine and X perks such as a checkmark. Your weekly usage pool covers Grok and Cursor's own Composer models only. Claude, GPT and Gemini come out of prepaid credit at API list price from the first request, and Teams and Enterprise add a $0.25 per million token Cursor Token Rate on most of them. OpenAI has also said it will end Cursor's direct access to its models on 12 November 2026. Cursor's chief executive, Michael Truell, says OpenAI models are about 5% of Cursor traffic and talks are continuing, but if your team runs on GPT, put that date in the calendar.
If you have to get it past a security team, the paperwork is in order. Cursor holds a SOC 2 Type II report and ISO 27001 and ISO 42001 certificates. Privacy Mode, on by default for teams and lockable by admins, keeps your code out of training by Cursor and its model providers. SAML or OIDC single sign-on and custom chat retention come with Teams; SCIM, audit logs and repository, model and MCP access controls need Enterprise.
Our take: a $30 Premium seat is a bargain if you are happy with Grok and Composer, and a quiet expense if what you really wanted was Claude Opus 5.5 at $4 in and $20 out per million tokens, plus the token rate. GitHub Copilot bills that same model from credits included in the seat price, with no X subscription attached, so it is worth pricing your own month on both before you commit a team.
Where it earns its place
A 25-developer product team on GitHub
Run agents on routine tickets and review every pull request automatically
Premium seats at $30 each cost $750 a month and include cloud agents and Bugbot. If the team works mostly in Grok 4.7 and Composer 2.5 the bill stays near that; switching to Claude Opus 5.5 adds API list price plus $0.25 per million tokens on every request.
A platform engineer at a fintech
Approve an AI editor without code leaking into model training
Cursor's help pages say Privacy Mode is on by default for team members and admins can lock it on; Cursor holds SOC 2 Type II and ISO 27001. Bring-your-own API keys fall outside Cursor's zero data retention agreements, and SCIM and audit logs need an Enterprise contract.
An engineering manager whose team relies on GPT models
Keep the same models after OpenAI's 12 November cutoff
Budget a migration to Claude, Gemini or Grok inside Cursor, or move GPT-heavy work to Codex or GitHub Copilot, which still lists GPT-6 models. Cursor says OpenAI models are about 5% of its traffic.
Before you commit
- Setup
- Download the editor; it imports VS Code settings and extensions in one step. Team checkout runs through grok.com.
- Integrations
- GitHub, GitLab and Bitbucket for Bugbot, Origin for code hosting, a CLI, and MCP servers, with admin controls over which MCP tools a team may use.
- Data and privacy
- Privacy Mode keeps code out of training by Cursor and its providers and is on by default for teams; SOC 2 Type II, ISO 27001:2022 and ISO 42001:2023, with regional data residency at a 10% uplift on eligible models.
- Cost at scale
- A 40-seat team on Premium pays $1,200 a month before third-party models; Claude Opus 5.5 then costs $4 per million input and $20 per million output tokens, plus a $0.25 per million Cursor Token Rate.
- Limits
- The included pool refreshes weekly and covers only Grok and Composer models; Plus cannot use third-party models, cloud agents or Bugbot at all.
What's good
- Agents Window runs many local and cloud agents side by side
- Cloud agents work in their own VMs and can test their own changes
- Weekly usage pool, so one heavy day does not burn the month
- SOC 2 Type II, ISO 27001 and ISO 42001; Privacy Mode enforceable by admins
- Requests are never downgraded when the pool runs out
What's not
- Claude, GPT and Gemini billed at API list price outside every plan
- Teams pay an extra $0.25 per million tokens on most third-party models
- OpenAI plans to end direct model access on 12 November 2026
- Sold only through X Subscription Corp, bundled with X and Grok perks
- SCIM and audit logs are Enterprise-only
Best for
Developers who want one VS Code-style editor for autocomplete, local agents, cloud agents and code review, and who are fine paying extra for non-Grok models.
Alternatives
Head to head
Cursor: common questions
- Is Cursor worth it?
- Still the editor where your agents can do the most, but since SpaceX bought it, every model that isn't Grok costs you extra. We rate it 4.5 out of 5.
- How much does Cursor cost?
- Free (limited); Plus $8/month (Cursor and Grok models only); Premium $30/month (4x Plus usage, cloud agents, Bugbot, third-party models at API list price); Super $100/month (15x); Ultra $200/month (50x); Teams $30, $100 or $200 per seat/month; Enterprise custom. Annual: $80, $300, $1,000, $2,000. Usage resets weekly.
- What are the downsides of Cursor?
- Claude, GPT and Gemini billed at API list price outside every plan; teams pay an extra $0.25 per million tokens on most third-party models; openAI plans to end direct model access on 12 November 2026; sold only through X Subscription Corp, bundled with X and Grok perks; sCIM and audit logs are Enterprise-only.
- Who is Cursor best for?
- Developers who want one VS Code-style editor for autocomplete, local agents, cloud agents and code review, and who are fine paying extra for non-Grok models.
- How much does Cursor cost for a team or in production?
- A 40-seat team on Premium pays $1,200 a month before third-party models; Claude Opus 5.5 then costs $4 per million input and $20 per million output tokens, plus a $0.25 per million Cursor Token Rate.
- Is Cursor safe for company data?
- Privacy Mode keeps code out of training by Cursor and its providers and is on by default for teams; SOC 2 Type II, ISO 27001:2022 and ISO 42001:2023, with regional data residency at a 10% uplift on eligible models.
- What are the best Cursor alternatives?
- The closest alternatives we have reviewed are Devin (incl. Devin Desktop, formerly Windsurf), GitHub Copilot, Claude Code and Zed.
Answers checked Oct 23, 2026.
Make Cursor? Show this rating on your site
