dalao.ai

Cursor

The coding agent for building ambitious softwareReference profile

About Cursor

Cursor is an AI coding environment from Anysphere, Inc. Its desktop editor is built on the open-source VS Code codebase (the docs say it is “based upon the VS Code codebase”), with an agent that reads the repository, edits files and runs commands, plus cloud agents, Bugbot reviews and Origin code hosting. The product itself is closed-source. Cursor announced on 14 Aug 2026 that SpaceX had acquired it. As of 2026-10-31, individual plans are Free $0, Plus $8, Premium $30, Super $100, Ultra $200 a month, sold as part of The X Subscription, with a weekly usage pool shared with Grok.

EditorialUpdated 13 min read

What problem it solves

Cursor sets out to let a coding agent work directly in the editor developers use every day, instead of copying code back and forth between a web chat box and the editor. The official migration docs say Cursor is "based upon the VS Code codebase", so extensions, themes, settings and keybindings can be imported in one click; the docs also say Cursor is regularly rebased onto newer VS Code versions, but for stability it often uses a "slightly older" VS Code version.

Where open source ends and closed source begins. The main VS Code repository, microsoft/vscode, is open source under the MIT license (LICENSE.txt in the repo). Cursor builds its own product on top of that open-source code, but Cursor itself is not open source: §5.1 of the Terms of Service says Anysphere and its licensors retain all rights in the service, and §1.5 prohibits reverse engineering, decompiling and creating derivative works. "Built on open-source code" and "closed product" therefore need to be understood separately.

Product surfaces (2026-10-25). The download page lists three entry points: the desktop app (macOS, Windows, Linux), a terminal CLI, and cloud agents launched from the browser, a phone, or Slack / GitHub / Linear. The recent changelog also shows: Projects (2026-09-10, beta; a coordinating agent splits the work and hands it to many sub-agents), two bots, Rollouts and Security Review (2026-09-23, Teams / Enterprise only), an iOS app for remotely viewing local agents (2026-10-06), the Origin App Marketplace (2026-10-13), and cloud-agent Canvases (2026-10-16).

Company and ownership. The contracting entity in the Terms of Service is Anysphere, Inc.; in §1.1, Anysphere describes itself as an applied research company working on automating coding. Funding disclosed on the official blog: a $60m Series A in 2024-08, a $105m Series B in 2025-01, a $900m Series C in 2025-06 (at a $9.9B valuation) and a $2.3B Series D in 2025-11 (at a $29.3B post-money valuation). The official blog announced a partnership with SpaceXAI in April 2026 to speed up model training, and on 2026-08-14 announced that "Cursor has officially been acquired by SpaceX", saying this completed the acquisition process that began when the SpaceXAI partnership was announced in April. The x.ai Terms of Service name SpaceXAI LLC as the contracting party and refer to it as "SpaceXSI"; Cursor's blog and model docs write SpaceXAI. The acquisition price was not disclosed. The Terms of Service (2026-09-03 version) are governed by the law of Texas, USA.

Current models (2026-10-25). The official model docs split models into two groups. "Cursor models" (Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5) count against the subscription's weekly usage pool. Third-party models (the docs name vendors such as OpenAI, Anthropic and Google, i.e. the Claude, GPT and Gemini families) don't count against the pool; on Premium and above they are charged at API list price from prepaid credit.

We still want to help people with ambitious ideas spend less time writing code and more time solving harder problems.

Cursor blog, “Cursor is now a part of SpaceX” (14 Aug 2026)

Core capabilities

The agent is the core. In the plan comparison table, Agent is available on every tier from Free up; cloud agents and automations, Origin code hosting, Bugbot review and third-party models all start at Premium. That makes Plus ($8/month) in practice an entry tier that "only uses Cursor's own models with fixed settings": the model docs say Plus has Fast turned off, effort fixed at medium and a 256k context for Grok 4.7, and does not support Auto, cloud agents, automations, Bugbot, Origin, third-party models or your own API key. Note: the pricing page says Plus includes "Only Grok models", while the model docs list Grok 4.7/4.6/4.5 and Composer 2.5; we follow the docs here. The changelog also says Projects are "Powered by Cloud Agents", which Plus does not include; Cursor doesn't say separately whether Plus can use Projects, and the changelog says that during the beta it is "rolling out to all users".

In-house models. Composer is Cursor's own agentic coding model. The Composer 2.5 blog post (2026-05-18) says that, like Composer 2, it is built on Moonshot's open checkpoint Kimi K2.5 and then trained with large-scale reinforcement learning, on 25 times as many synthetic tasks as Composer 2 (vendor claim). The Composer page says it can be used only in Cursor (desktop, web, CLI, mobile, plus the official SDK). The model docs call Grok 4.7 "a frontier model from Cursor and SpaceXAI" and Grok 4.5 "a joint model from Cursor and SpaceXAI". Grok 4.7 was released on 2026-09-21; the official blog calls it "our most capable model for long-running coding and knowledge work" (vendor claim) and links to the full announcement on x.ai.

Own-model rates. Beyond the weekly pool (2026-10-25, per million tokens), Grok 4.7: input $2, cache read $0.5, output $6; Fast $4 / $1 / $12 (2×). The Grok 4.7 doc says that when input exceeds 256k tokens, standard requests are priced at 2× and Fast requests at 3× the standard price, with a maximum context of 500k, and only input length decides whether this applies. Composer 2.5: input $0.5, cache read $0.2, output $2.5; Fast $3 / $0.5 / $15 (input and output about 6× the standard price).

How third-party models are billed. The docs' model price table lists input, cache write, cache read and output prices for each third-party model and flags whether it supports US data residency. For example, Claude Opus 5.5 is $4 input and $20 output, and GPT-5.6 Terra is $2 input and $12 output. On eligible third-party models, Teams and Enterprise also pay a "Cursor Token Rate" of $0.25 per million tokens on top of the API list price (BYOK usage included; a set of small models such as Claude Haiku 5.5 and GPT-5 Mini is exempt); individual plans don't pay this fee.

Auto routing. Auto has three modes: Cost, Balance and Intelligence. Individual plans get only Cost mode, billed as Cursor models and drawn from the weekly pool; on Teams / Enterprise, Cursor Router chooses between Cursor models and third-party models for each Balance or Intelligence request, and requests routed to a third-party model are billed as third-party usage.

How it compares

Versus VS Code itself. The main VS Code repository's code is open source under the MIT license. Cursor is built on that code, and extensions, themes and settings migrate in one click. The difference is that Cursor is a closed, standalone application whose AI features are tied to a Cursor account and Cursor's servers: the data-use page says that even when users enter their own API key, requests still pass through Cursor's backend, where the final prompt is assembled.

Versus GitHub Copilot. Prices are as of 2026-10-25, from GitHub's official pricing page. Copilot's individual plans are Free $0, Pro $10/month, Pro+ $39/month and Max $100/month, metered in GitHub AI Credits (1 credit = $0.01) that reset monthly; code completions don't consume credits. Cursor's individual plans are Free $0, Plus $8, Premium $30, Super $100 and Ultra $200/month, with a usage pool that resets weekly and is shared with Grok. In form, Copilot comes as an extension in many editors, including VS Code, JetBrains, Xcode and Neovim; Cursor is a standalone editor plus a CLI and cloud entry points. The two meter usage in different units (monthly credits versus a weekly pool plus third-party models at API list price), so monthly fees alone can't tell you "which is cheaper".

Training defaults. The FAQ on GitHub's pricing page says interaction data from Copilot individual users (Free / Pro / Pro+) may be used to train models, with an opt-out in settings. Cursor's Privacy Mode is available to everyone; when it is off, Cursor may use code data to train models (data-use page). Both require users to check their settings themselves.

Model sourcing. The included allowance in a Cursor subscription covers only its own models (the Grok series and Composer); using Claude, GPT or Gemini costs extra at API list price. If a team relies mainly on third-party frontier models, the included allowance is of limited use and actual spend comes mostly from prepaid credit.

Who it's for / Not for

Who it's for

Developers already using VS Code who want to bring their extensions and habits over unchanged, and want an agent to edit multiple files and run commands directly in the editor.

Individuals willing to use Cursor's own models (Grok 4.x, Composer 2.5) as their main models and cover daily work with the weekly usage pool; when they need third-party models, they accept paying API list price from prepaid credit.

Teams that need cloud agents, Projects, Bugbot review and central administration: Teams costs $30 / $100 / $200 per seat, and Enterprise offers SCIM, audit logs and fine-grained spend controls, with Privacy Mode on by default for Enterprise teams.

Enterprises with US data-residency needs: Enterprise can enable in-US data residency per team, with a 10% uplift on eligible models.

Not for

Teams that require both the editor and the AI backend to be open source, self-auditable or self-hostable: Cursor is a closed commercial service, and its terms prohibit reverse engineering.

Organizations that can only use their own model gateway and require that no request pass through the vendor's backend: Cursor says that even with your own key, requests go through Cursor's backend.

People who subscribe only to Plus but want Claude, GPT or Auto: Plus can only use Cursor's own models, with fixed settings.

Users in countries or regions under US embargo: §17.5 of the Terms of Service prohibits use there.

Enterprises that need EU or APAC data residency: as of 2026-10-31 only US residency is available, and EU plus UK is offered only at the inference layer, on application.

Getting started

Install. Get the macOS / Windows / Linux desktop app from the download page; terminal users can install the CLI with the official install script (curl https://cursor.com/install -fsS | bash). On first launch, click "VS Code Import" under Cursor Settings → General → Account to import extensions, themes, settings and keybindings in one go.

Turn on Privacy Mode first. Privacy Mode is turned on in settings. Team admins can enable and enforce it at the team level so members can't turn it off; enterprises can also use the MDM Allowed Team IDs policy to block signing in to personal accounts on company devices.

Choose a tier. Third-party models require at least Premium; prepaid credit, auto top-up and the monthly additional-usage cap are set on the billing page. The weekly usage pool's remaining amount and reset time are shown both in the editor settings and under Usage at grok.com/dashboard. Annual prices are Plus $80, Premium $300, Super $1,000 and Ultra $2,000 a year (excluding tax).

Bring in a team. Teams is bought per seat and can mix Premium, Super and Ultra seats; Plus can't be used for teams. SSO (SAML/OIDC) is available from Teams up; SCIM, invoice payment, audit logs, and model and MCP access controls require Enterprise.

Check before using cloud agents. Cloud agents are the only feature that needs Cursor to store code: an encrypted copy of the repository is kept only while the agent runs and is deleted when the task finishes. Teams whose security policy forbids storing code externally are not a fit for cloud agents; other features are unaffected.

Prices and model figures are all as of 2026-10-25, from cursor.com/pricing and the model docs.

Pricing & license

Prices as of 2026-10-25, from cursor.com/pricing and the official Models & Pricing docs. All prices are in USD, excluding tax.

Individual subscriptions (monthly). Free $0 (limited Agent and Chat); Plus $8 (basic usage, Cursor's own models only); Premium $30 (4× Plus usage); Super $100 (15×, priority at peak times); Ultra $200 (50×, first access to new products and models). Annual: Plus $80, Premium $300, Super $1,000, Ultra $2,000 a year. The pricing page says subscriptions are sold only directly by X Subscription Corp, with no authorized resellers; the pricing page also lists X platform and Grok benefits.

Teams and Enterprise. Teams costs $30 / $100 / $200 per seat per month, with usage matching the individual tier of the same level; Enterprise is custom-priced with pooled usage.

Usage billing (separate from the subscription). The subscription includes a weekly usage pool shared by the Cursor editor, CLI, cloud agents and Grok apps, and different models draw it down at different rates. When it runs out there are three options: wait for next week's reset, continue on prepaid credit at per-model prices, or upgrade; Cursor says requests are not degraded in quality or speed as a result. Third-party models are charged at API list price from prepaid credit from the very first request and are available only on Premium and above. Teams / Enterprise pay an extra $0.25 per million tokens on most third-party models. US data residency adds 10% on eligible models. The Grok API is billed separately at standard rates and doesn't use the pool.

Legacy plans. Existing customers on Pro, Pro+, Ultra, Start and the Teams Standard / Premium plans keep their original rules: two monthly usage pools plus on-demand billing, with no automatic migration.

License. Cursor is closed-source commercial software (Terms of Service §1.5 and §5.1); the main VS Code repository code it is built on is MIT-licensed. In-house models such as Composer 2.5 are offered only inside Cursor, with no public weights. Subscription fees are non-refundable except where the law requires otherwise, and cancellation must be made at least 24 hours before renewal (Terms of Service §4.1 and §4.4).

Risks & limits

The usage pool is shared across products. The weekly usage pool is shared by Cursor, Grok Chat, Imagine, Voice and Grok Bot. If the same account generates a lot of images or video in Grok, that crowds out the allowance for coding.

Third-party model costs have no ceiling. Claude, GPT and Gemini are all charged at API list price from prepaid credit. Long context, Fast mode (about 2× for Grok, about 6× for Composer 2.5 Fast), US data residency (+10%) and the team Token Rate stack up. Without a monthly additional-usage cap, this spend has no ceiling once auto top-up is on.

Privacy Mode exceptions. Even with Privacy Mode on, model providers (Cursor included) may still run risk classifiers, and conversations that trigger abuse detection may be retained for investigation. Some models (such as Claude Fable 5 / 5.1) are outside the zero-data-retention agreements; under Privacy Mode or on enterprise accounts, an admin must first approve data retention before they can be used. When a Fable request trips its safety guardrails, Cursor automatically reroutes it to Claude Opus. With Privacy Mode off, Cursor may store code data and use it to train models.

Requests always go through the vendor's backend. Even your own key doesn't bypass Cursor's backend; file contents are cached temporarily, encrypted with client-generated keys. US data residency doesn't cover BYOK, SSO (via WorkOS), user-configured MCPs or @Web.

What region and compliance mean in practice. The Terms of Service require users to comply with applicable trade laws and prohibit using the service in US-embargoed countries or regions. The security page says Cursor uses or maintains no infrastructure in China and uses no Chinese companies as subprocessors. The Terms of Service restrict use in some regions and for some purposes; the original text governs. Disputes are governed by Texas law, with litigation in Wichita or Tarrant County in that state.

Change of ownership. Since the SpaceX acquisition on 2026-08-14, pricing, terms and the selling entity have been tied to Grok and X (X Subscription Corp). The terms let Anysphere modify the service at any time, possibly without separate notice (§10); fee changes require advance notice (§4.2).

Auto-run code. §1.7 of the terms says that once auto-run is turned on, the risks (data loss, security vulnerabilities and so on) are borne by the user.

Vendor figures. Revenue and customer numbers (for example, the claim in 2025-11 of annualized revenue above $1 billion) and descriptions of model capability are all vendor claims.

Our verdict

Cursor's strength is "a familiar VS Code base plus a deeply integrated agent": migration costs little, and cloud agents, Projects, Bugbot and Origin connect coding, review and hosting into one line, all of it written up in the official docs and changelog. The 2026 changes are just as clear: after the SpaceX acquisition, its own models gained the Grok series launched jointly with SpaceXAI (Composer 2.5 is still on sale), and the subscription was folded into The X Subscription, with a usage pool shared with Grok.

The choice depends on three things. First, whether your main models are Cursor's own: if so, the weekly pool on Premium or Super is the best value; if you mainly use third-party models, the subscription is just the entry ticket and most of the money goes to prepaid credit. Second, the data boundary: Privacy Mode is off by default, every request passes through Cursor's backend, and cloud agents involve temporary storage of code. Third, the product is closed source, and its direction may shift with its parent company.

Our verdict: Cursor suits individuals and teams already working in VS Code who need strong agent capabilities; it doesn't suit organizations that require open source, self-hosting or data residency outside the US.

Why it matters

Cursor built the coding agent into the editor engineers use every day.

Sources

Official

  1. Cursor · Pricingcursor.com · Accessed Oct 25, 2026
  2. Models & Pricingcursor.com · Accessed Oct 25, 2026
  3. Cursor · Securitycursor.com · Accessed Oct 25, 2026
  4. Cursor · Data Use & Privacy Overviewcursor.com · Accessed Oct 25, 2026
  5. Cursor · Privacy Policycursor.com · Accessed Oct 25, 2026
  6. Privacy and Data Governancecursor.com · Accessed Oct 25, 2026
  7. Cursor · Terms of Servicecursor.com · Accessed Oct 25, 2026
  8. VS Code Migrationcursor.com · Accessed Oct 25, 2026

Updates & reading