Cursor vs Windsurf: what published pricing and docs actually say

The short answer

On published prices, the two ladders barely overlap. Cursor's individual plans run $0 / $8 / $30 / $100 / $200 per month, with paid tiers sized as multiples of the $8 plan's usage. Devin Desktop — the product windsurf.com now redirects to — runs $0 / $20 / $200 per month for individuals, plus a Teams plan at $80/month for the team and $40/month per full developer seat.

The structural difference is what the entry price buys. Cursor's Plus plan at $8 includes only Cursor's first-party models (the Grok family and Composer). Devin's Pro plan at $20 publishes frontier-model access from OpenAI, Claude, Gemini and SpaceXAI plus leading open-source models, and free use of Cognition's own SWE-2 model. Neither vendor publishes an absolute request or credit count on its pricing page — Cursor expresses allowances as multipliers, Devin as qualitative quota language.

Two products, two new owners

Any current comparison has to start with ownership, because it explains both pricing pages. Cognition announced on 14 July 2025 that it had signed a definitive agreement to acquire Windsurf — its IP, product, trademark and brand — citing $82M of ARR and 350+ enterprise customers at the time. On 2 June 2026, Cognition folded the editor into its Devin brand: Windsurf became Devin Desktop, delivered as an over-the-air update. The company states that plans, pricing, extensions and features remain the same, that the classic Windsurf IDE is still fully accessible alongside the new Agent Command Center, and that the editor remains backwards-compatible with Windsurf and VS Code. Confirming the transition in practice, windsurf.com/pricing now redirects to devin.ai/pricing, and the windsurf.com homepage redirects to devin.ai/desktop.

The rebrand also replaced the product's local agent. Cognition documents Devin Local as the successor to Cascade, rewritten in Rust and claimed to be up to 30% more token efficient — a vendor figure, not an independently checked one. Devin Desktop also launched with support for the Agent Client Protocol (ACP), so third-party agents including Codex, Claude Agent and OpenCode can run inside the editor.

Cursor's change is commercial rather than nominal. Cursor's help documentation now states that Cursor is part of The X Subscription, one subscription covering Grok, Cursor and X, with new customers choosing an X Subscription plan and billing managed at grok.com/dashboard. The pricing page FAQ goes further: subscriptions are "only sold directly through X Subscription Corp", with no authorized resellers. The site footer still reads © 2026 Anysphere, Inc. Secondary reports of a $60 billion all-stock acquisition of Anysphere by SpaceX could not be confirmed from reputable coverage fetched for this article; see what we could not verify.

Cursor's published plans and prices

The X Subscription ladder published on cursor.com/pricing and in Cursor's help documentation:

• Free — $0. Limited AI usage: limited agent requests and limited chat requests, using Cursor's first-party models.
• Plus — $8/month ($80/year). A "Basic" shared AI usage pool, Grok Chat, Voice, Imagine and Build, and Grok API access (API usage billed separately at standard model rates). Plus runs Cursor models at fixed settings — Fast off, medium effort — and excludes Auto routing, Cloud Agents, Automations, Bugbot, Origin and third-party models. It bundles X features such as the premium checkmark and fewer timeline ads.
• Premium — $30/month ($300/year). 4× Plus usage, latest Grok models, cloud agents and automations, code hosting via Origin, AI code reviews via Bugbot, Grok Bot, and third-party models at API list price paid from prepaid credit.
• Super — $100/month ($1,000/year). 15× Plus usage and priority model access at peak times.
• Ultra — $200/month ($2,000/year). 50× Plus usage and earliest access to all products and models.

Usage mechanics are documented in detail: every paid plan includes one weekly pool of usage for Cursor's own models — Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5 — shared across Cursor and the Grok apps. The pool refreshes weekly, unused allowance does not roll over, and overage draws on prepaid credit that currently expires 365 days after purchase. Prices exclude tax, and the pricing page warns that third-party model usage is never included in the monthly price.

Devin Desktop's published plans and prices

devin.ai/pricing (where windsurf.com/pricing now points) publishes a shorter ladder:

• Free — $0. A light quota to code with agents, limited model availability, unlimited inline edits and unlimited Tab completions.
• Pro — $20/month (labelled "popular"). Increased quotas including access to frontier models from OpenAI, Claude, Gemini and SpaceXAI plus leading open-source models; full model availability; free use of SWE-2; access to cloud agents (Devin Cloud); and the option to purchase extra usage at API pricing.
• Max — $200/month (labelled "new"). Everything in Pro plus "significantly higher quotas" — the pricing page's own wording, with no number attached.
• Teams — $80/month for the team plan plus $40/month per full dev seat. Up to 200 users, sharing and collaboration, centralized billing, an admin dashboard with analytics and priority support. Each full seat includes its own quota and access to Devin Desktop.
• Enterprise — custom. SAML/OIDC SSO, centralized enterprise admin controls, a dedicated deployment option (including deploying in your virtual private cloud), highest-priority support and dedicated account management.

Devin's usage FAQ says each paid plan's allowance "refreshes automatically on a daily and weekly basis", that cost per message varies with the model, task size and reasoning required, and that overage is consumed at API pricing. It even publishes cost-management advice: switching to smaller models such as Haiku, GPT 5.2 Mini or Kimi K2.5 extends the included usage.

Usage allowances, side by side

The headline comparison, entirely from the vendors' published pages and docs:

Read literally, these are different commercial structures. Cursor meters a weekly pool sized by tier and treats every third-party model as pass-through billing. Devin bundles third-party frontier models into the $20 plan and meters an unnamed quota that refreshes daily and weekly. A buyer comparing "included usage" across the two has no common unit: neither vendor publishes request counts, token budgets or a dollar value for its included allowance.

Model support and platforms

Models. Cursor's documentation splits the catalogue into first-party models covered by the included pool (Grok 4.7, Grok 4.6, Grok 4.5, Composer 2.5) and third-party frontier models from OpenAI, Anthropic and Google, unlocked on Premium and above at published API rates with spending caps. Free and Plus are restricted to the first-party set. Devin's pricing page advertises "first class support for every major model provider" and names OpenAI, Claude, Gemini and SpaceXAI frontier models plus leading open-source models on Pro, with SWE-2 — Cognition's own coding model — included free.

Platforms. Cursor's download page offers macOS builds (Apple Silicon, Intel and universal), Windows installers for x64 and ARM64 in both user and system variants, and Linux packages as .deb, .rpm and AppImage for x64 and ARM64. Devin Desktop's install documentation lists Mac (minimum OS X Yosemite), Windows (minimum Windows 10), and Linux via tarball, apt or yum repositories — requiring glibc 2.28 or newer, which covers Ubuntu 20.04+, Debian 10+, Fedora 36+ and RHEL 8+. Both products therefore cover the three major desktop operating systems; neither pricing page publishes mobile editor apps, though Cursor's plan materials reference spawning cloud agents from a browser or phone.

Teams and enterprise pricing

Cursor prices Teams per seat, mirroring the individual tiers: Premium seats at $30/user/month, Super at $100 and Ultra at $200, with each seat carrying the matching weekly usage pool and a shared prepaid balance for overage. The documentation notes that requests made with your own API key on Teams and Enterprise still carry a Cursor Token Rate of $0.25 per million tokens. Enterprise adds pooled usage, SCIM provisioning, audit logs, advanced admin controls and invoice billing, at custom pricing.

Devin splits the bill differently: a flat $80/month team plan plus $40/month per full developer seat, supporting up to 200 users, with centralized billing, an admin dashboard and priority support included. Enterprise terms are custom. Cursor's per-seat model scales linearly with tier choice; Devin's per-seat price is fixed but its quota contents are not quantified.

What we could not verify

• Real-world completion quality. DealParse has no lab. Claims like Cognition's "up to 30% more token efficient" for Devin Local, or Cursor's model-by-model guidance, are vendor statements we can quote but not check. Neither vendor publishes benchmark results on the pages we fetched. Not verified
• Absolute usage allowances. Cursor defines tiers only as multiples of an unquantified Plus pool; Devin uses "light", "increased" and "significantly higher" without numbers. No published request counts, token budgets or dollar-equivalents exist on either pricing page. Not verified
• The reported SpaceX acquisition of Cursor's parent. Reputable coverage of a reported $60 billion all-stock acquisition of Anysphere (Forbes, Reuters, CNBC) was inaccessible to us this session behind bot walls and cookie blocks. What Cursor's own documentation confirms is narrower: Cursor is part of The X Subscription and subscriptions are sold only by X Subscription Corp. The deal's size and structure remain unconfirmed here. Not verified
• Enterprise pricing on both sides. Both vendors publish "custom" / "let's talk" for enterprise agreements; no rate cards are public.
• Cross-vendor comparability of quotas. Because the units differ, no published conversion lets a buyer equate one plan's allowance with the other's.

Prices and plan structures in this space have changed twice within eighteen months across these two vendors alone. This analysis reflects what Cursor and Cognition published on 19 October 2026; anything not on those pages is labelled unverified rather than presented as fact.