Cursor Review (September 2026): OpenAI Models Leave November 12, and the Rating Moves

On August 28, OpenAI told SpaceX it intends to wind down the contract that puts its models inside Cursor, with a proposed shutoff of November 12, 2026, per OpenAI. Cursor CEO Michael Truell puts the blast radius at roughly 5% of Cursor traffic, per Engadget. So the editor millions of developers open every morning is losing one model lane, not its spine. The rating moves from 4.4 in August to 4.2, and the reason is closer to governance than to capability: a frontier lab just said in public that it does not trust the company that now owns your editor.

Here is the short answer before the detail. Solo devs on Pro: keep it, and move your Tab model off GPT if you were on it. Small teams: do not sign an annual Teams contract until the post-split model terms are written down. Enterprise: require a model-access guarantee in the contract, and pin your models with Cursor's own Enterprise controls. The reasoning runs tier by tier below, with pricing dated to today.

What the OpenAI termination actually affects

The thing being switched off is narrow and specific: OpenAI's GPT models as a selectable option inside Cursor's Composer, Chat, and Tab. OpenAI gave the maximum notice its contract allows, which is why the date is November 12 rather than next week, per OpenAI. Cursor can choose an earlier cutoff, and the two companies are still talking, with Truell saying his team is speaking with OpenAI in hopes of a resolution, per Engadget.

Read the 5% figure honestly and it reframes the whole story. If OpenAI models account for about one in twenty Cursor sessions, most teams already run Claude or Gemini as their primary model and will not notice the switch flip. The developers who feel it are the ones who deliberately pin GPT for a specific job: a particular Composer refactor pattern, or GPT-driven Tab completions in a language where they prefer its suggestions. That is a real group, and it is a minority.

One loss is sharper than the wind-down suggests. OpenAI said it will hold the cancellation to the latest date it can "while not providing future models to Cursor," and it named its upcoming Astra model directly, per OpenAI. Even during the grace period, the top of the OpenAI lane is already closed. If Astra ships before November 12 and you want it in an editor, Cursor will not be where you get it.

Why is OpenAI doing this? Its stated reason is that it cannot be confident SpaceX will use the models within its terms of service, citing Twitter breaking a prior OpenAI contract and Elon Musk's admission under oath that xAI distilled OpenAI training data, per OpenAI. Whatever you make of the feud, the mechanism matters for buyers: SpaceX closed its acquisition of Anysphere, Cursor's maker, in mid-August, and a change of control gave OpenAI a contractual window to walk. The ownership change is what opened the door.

The model lineup after November 12

Strip out OpenAI and Cursor still ships with three frontier families on the menu. Claude from Anthropic, Gemini from Google, and Grok from xAI all remain selectable, and the product page confirms access to frontier models on the paid tiers, per Cursor. Grok is the interesting one here, because SpaceX now owns xAI, so the model most likely to become Cursor's cheap default is the one its parent builds. Anthropic is moving the other way: Tom Brown said the company will increase compute to support Claude models in Cursor, per Engadget, a Claude-first signal that the default most teams already run gains headroom just as the OpenAI lane closes.

Model choice has always been the marketed feature, and the changelog says the editor around it is not slowing down. In the two weeks before this review, Cursor shipped Origin code hosting on August 17, expanded cloud agents on August 19, and added start-from-scratch agents that spin up a repo on August 27, per Cursor. There is no feature freeze and no post-acquisition stall. The team that earned the rating is still shipping at pace.

So the capability question is not "does Cursor still work." It does. The question is whether a four-supplier editor quietly becoming a three-supplier editor, owned by a company that builds the fourth model, is a direction you want to be locked into. That is a governance judgment, and it is exactly where the half-tenths of the rating live.

Cursor versus Claude Code, and who should actually switch

The reflex after a headline like this is to search for an exit, and the name that comes up most is Claude Code. Reach for it as a Cursor replacement and you will pick up the wrong tool. Claude Code is a terminal program, not an IDE. It runs an agent loop over your repository from the command line; it does not give you inline Tab completions, a visual diff-and-apply Composer surface, or Bugbot-style review inside an editor window. Anthropic prices Claude Code access from the $20/mo Pro plan, with heavier use on Max from $100/mo, per Anthropic. It is excellent at repo-scale agentic work driven from a shell, and it locks you to one model family by design.

If you want an IDE and you are shopping because of the OpenAI split, the honest comparisons are Windsurf and GitHub Copilot, not Claude Code. Copilot keeps GPT, Claude, and Gemini all on the menu, and its individual tiers run $10/user/mo for Pro and $39/user/mo for Pro+, per GitHub. For a team already living inside GitHub, the native pull-request review and org billing are the reasons to look, and the multi-model access Cursor is losing on one lane, Copilot keeps on all of them.

Here is the three-way decision table, pricing dated 2026-08-31.

Tool	Category	Models available	Entry pricing	Best fit
Cursor	IDE (VS Code fork)	Claude, Gemini, Grok (OpenAI out Nov 12 barring a deal)	Pro $20/user/mo; Teams $40/user/mo, per Cursor	IDE-first devs who want multi-model choice with Tab and Composer in the editor
Claude Code	CLI (terminal)	Claude only	From $20/mo Pro; Max from $100/mo, per Anthropic	Repo-scale agentic tasks from the shell; teams committed to Anthropic
GitHub Copilot	IDE plugin (VS Code, JetBrains)	GPT, Claude, Gemini	Pro $10/user/mo; Pro+ $39/user/mo, per GitHub	Teams already in GitHub who want native PR review and broad model access

Read across the rows and the switching logic is clean. Move to Claude Code only if you were going to live in the terminal anyway; it is a category change, not a like-for-like swap. Move to Copilot if GitHub is already your home and multi-model access is the thing you refuse to give up. Stay on Cursor if you want an IDE-first workflow with Claude or Gemini as your driver, because that is precisely the setup the OpenAI exit does not touch.

Cursor pricing, as of 2026-08-31

Prices pulled live from the pricing page today, so treat these as current and the August figures as stale. Every paid tier includes a set amount of model usage and then bills on-demand in arrears, per Cursor.

Plan	Price	What you get
Hobby	Free	Limited agent requests and Tab completions, per Cursor
Individual (Pro)	$20/user/mo	Extended agent limits, frontier models, cloud agents, per Cursor
Teams	$40/user/mo	Everything in Individual plus centralized billing, SSO, and team-wide privacy mode, per Cursor
Enterprise	Custom	Pooled usage plus repository, model, and MCP access controls, per Cursor

The Individual card carries Pro, Pro+, and Ultra tiers behind a toggle, and Teams splits into Standard and Premium at higher usage, per Cursor. One line item earns its keep this month: the Enterprise plan lists "repository, model, and MCP access controls" as a feature. That control is now the thing an enterprise buyer should actually use, because it lets an admin pin which models a team can reach rather than trusting a verbal promise about them.

The pricing page also offers a monthly-versus-yearly toggle. Yearly is cheaper per seat, which in a normal month is the smart buy. This is not a normal month, and the discount is the trap discussed below.

The verdict, by who you are

Split it by buyer, because the OpenAI exit lands on each one differently.

Solo developer on Pro at $20/mo, per Cursor: keep Cursor. If Claude or Gemini already drives your work, November 12 changes nothing you will feel. If you specifically pinned GPT for Tab completions, switch your Tab model to Claude Sonnet or a Gemini Flash tier before the deadline and judge the difference on your own code; for most autocomplete work the gap is small. The switching cost off an IDE you know outweighs the loss of a 5% model lane. Stay month-to-month so your exit stays clean if the direction sours.

Small team of 3 to 10 on Teams at $40/user/mo, per Cursor: do not sign the annual contract yet. The yearly discount looks like free money, and in a stable market it would be. Right now a supplier just walked, and the post-split model terms are not written down. Pay the month-to-month rate and treat the premium over annual as insurance against a second provider following OpenAI out the door. When Cursor publishes a written commitment on which models stay and how your code is routed, re-run the math and lock in annual then, not before.

Enterprise buyer evaluating 50-plus seats: deployment is fine, but make model access contractual. In August the risk was a verbal "nothing is changing" from a company that also builds Grok; now that risk has a live example attached, because OpenAI just demonstrated that a supplier can leave over a trust dispute with the owner. Require a written model-access and data-routing guarantee before you renew, and put Cursor's own Enterprise model-access controls to work pinning the exact models your teams depend on, per Cursor. If SpaceX will not commit the model roster in writing, that refusal is your answer.

A note on the number. The 4.4 in August priced the OpenAI question as an unresolved risk. It is resolved now, in a specific direction: one model lane closes, the frontier OpenAI models are already off the table, and a major supplier publicly distrusts the owner. Against that, the editor is undegraded and shipping weekly, and the roughly 95% of usage that is not OpenAI-based is untouched, the flip side of the 5% traffic figure Cursor's CEO gave, per Engadget. Two tenths down to 4.2 is the honest weight of a narrow capability loss plus a governance flag, not a collapse. If a deal closes before November 12 and Astra-class access returns, the number goes back up.

What to do this week

Solo devs: open Cursor settings and check which model your Tab and Composer are set to. If it is a GPT model, switch it to Claude or Gemini today and spend the week judging the swap on your own repo, so November 12 is a non-event instead of a surprise.

Small teams: pull your Teams billing page and confirm you are on monthly, not annual. If a renewal is due before mid-November, hold it at month-to-month and put a calendar note to re-decide once Cursor publishes post-split model terms.

Enterprise buyers: send your Cursor or SpaceX account contact one email this week asking for a written commitment on model availability and data routing after the OpenAI wind-down, and ask your admin to configure Enterprise model-access controls around the models your teams actually use. The answer you get back, or the silence, tells you whether to renew.

If you are deciding fresh and want an IDE-first editor with Claude or Gemini as the driver, Cursor is still the pick; start on the free Hobby plan to judge the Tab model on your own language before you pay, then move up from Cursor's site if it earns the seat. The multi-model workflow that made Cursor worth recommending is the part the OpenAI exit leaves standing.