Cursor Models Are About to Change: OpenAI Pulls Out on 12 November

TL;DR: OpenAI is ending its deal with Cursor following SpaceX’s takeover, and has proposed 12 November 2026 as the date Cursor loses direct access to OpenAI’s models. The Cursor models list keeps Claude, Gemini, Grok and Cursor’s own Composer — it loses GPT. Anthropic has publicly committed to increasing Claude capacity in Cursor, so the models most people actually use are staying. If you build in Cursor and never touch the model picker, the honest answer is that you probably won’t notice, and you have ten weeks either way.

What happened

On 28 August 2026 OpenAI published a statement titled Our decision on Cursor following its acquisition by SpaceX, describing its decision to wind down the contract that supplies OpenAI models to Cursor. Reuters reported the story on 29 August, and it went straight to the top of Hacker News.

The specifics, per Reuters:

Proposed shutoff date	12 November 2026
Mechanism	A change-of-control clause in the existing contract
Reason given	OpenAI “cannot be confident that SpaceX will use our technology within our terms of service”
Status	Not final — Cursor says it is still talking to OpenAI

OpenAI’s stated reasoning is about trust rather than performance. Its statement, as quoted by Reuters, is that it is “making this choice because we cannot be confident that SpaceX will use our technology within our terms of service, based on our experience with Elon Musk’s companies violating contracts” — a reference to X, now part of SpaceX, which OpenAI says breached a contract after Musk acquired it.

Cursor is not treating it as settled. Michael Truell, Cursor’s co-founder and now a SpaceX executive, said the company was “speaking with the OpenAI team to resolve this issue”, adding that Cursor had “trusted their platform to be neutral infrastructure for our business”. Musk’s own response, per Reuters, was rather shorter: “I couldn’t care less.”

This follows SpaceX’s $60bn acquisition of Cursor closing earlier this month — the consequence we said to watch for, arriving faster than expected.

Which Cursor models actually go away

This is the part the wire coverage skips, and it’s the only part that changes your day.

Per Cursor’s own documentation, the Cursor models you can pick today span five providers. Here’s what 12 November does to that list:

Provider	Models in Cursor	After 12 November
OpenAI	GPT-5.5, GPT-5.6 Sol	Gone, under OpenAI’s proposal
Anthropic	Claude Opus, Claude Opus 5, Claude Fable 5	Unaffected
Cursor’s own	Grok 4.5, Composer	Unaffected
Google	Gemini Pro	Unaffected

Two details matter more than the table. First, Cursor’s default setting is Auto, which picks a model for you — and Cursor’s docs list GPT-5.5 as one of the models Auto routes to. So if you have never opened the model picker, OpenAI models may well have been doing some of your work without you choosing them. After 12 November, Auto simply routes elsewhere.

Second, Claude isn’t going anywhere — and the reason is worth understanding, because it’s the opposite of reassuring in the abstract while being genuinely reassuring in practice.

Why Claude is staying (and what that tells you)

Within hours of OpenAI’s statement, Anthropic co-founder and chief compute officer Tom Brown posted publicly that Cursor “has been a trusted partner of Anthropic since Sonnet 3.5”, and that Anthropic will “continue to increase compute to support Claude models in Cursor”.

That is a clear commitment, and it matters more to you than OpenAI’s departure does — Claude models are what most people actually run in Cursor. But as The New Stack reported on 31 August, it sits oddly against Anthropic’s own recent history. Anthropic cut Windsurf off from Claude in 2025 on nothing more than rumours that OpenAI might buy it, and cut xAI staff off in January.

So why is a company now wholly owned by an Anthropic rival still a “trusted partner”? The likely answer is money flowing the other way. Anthropic buys enormous amounts of computing power from SpaceX — per SpaceX’s IPO filing, around $1.25 billion a month, running to 2029, for the data centres that keep Claude’s usage limits where they are.

Replit’s chief executive Amjad Masad said the quiet part publicly: “Maybe you changed your ways, but we all remember what you did to Windsurf… More likely answer is that you can’t do that here because you need the compute.” Engineer Gergely Orosz made the same point — SpaceX is “in this enviable position where one of its biggest competitors depends on its compute infra”.

For you, the practical read is straightforward: Claude in Cursor is on much firmer ground than GPT was, because the commercial relationship runs in Anthropic’s disfavour, not its favour. The less comfortable read is that this was never about which tool is best. Your model line-up is the output of contracts you can’t see and didn’t sign.

Why it matters if you don’t code

Here is the plain version: you are renting someone else’s models through someone else’s app, and neither relationship is yours.

Nothing about your projects breaks. Your apps keep running, your files stay where they are, and nothing needs migrating. What changes is the list of Cursor models on the menu.

• If you use Auto (most non-developers do): nothing to do. The quality of Cursor models may shift slightly as Auto stops reaching for GPT, and you may not be able to tell.
• If you deliberately pick GPT-5.6 Sol for certain jobs: that option has roughly ten weeks left. Start trying Claude Opus 5 or Composer on those same jobs now, while you can still compare them side by side.
• **If you chose Cursor because of GPT:** this is a genuine reason to reconsider, and you have time to do it calmly.

There’s a broader lesson worth taking, and it isn’t about Cursor. Every AI builder — Cursor, Lovable, Replit, Bolt — is a wrapper around models it doesn’t own. Corporate fallings-out you have no visibility into can remove a model from your toolbox with ten weeks’ notice. The defence isn’t picking the “right” tool; it’s not building a workflow that only one model can run.

What to do in the next ten weeks

1. Check which of the Cursor models you’re actually using. Open the model picker in Cursor. If it says Auto, you’re in the low-effort majority and this is mostly noise.
2. If you use GPT deliberately, run a comparison this week. Same prompt, same project, GPT-5.6 Sol versus Claude Opus 5 versus Composer. Ten minutes now beats a scramble in November.
3. Don’t cancel anything yet. The date is a proposal in a live negotiation, not a switch already flipped. Truell says talks are ongoing.
4. Write down your two or three most valuable prompts. Portable prompts are the thing that survives this kind of news. Tool loyalty isn’t.
5. If you’re mid-project, carry on. Ten weeks is a long time, and nothing stops working before then.

Who should care (and who shouldn’t)

• Building a revenue app in Cursor: worth thirty minutes this week to confirm you’re not dependent on one model. Then get back to building.
• Building an internal tool: ignore this. Auto will handle it.
• Building a site or landing page: irrelevant — you’re not model-shopping for that.
• Learning, or still choosing a tool: don’t let this scare you off Cursor, but do let it inform how you choose. If you’re weighing your options, the 60-second quiz is a better guide than a headline about a corporate row.
• Already on Claude, Lovable or Replit: the Cursor models list is not your problem today. File it under “how this industry works”.

Our take

We’d not switch tools over this, and we’d be wary of anyone telling you to. Cursor keeps Claude, Gemini, Grok and its own Composer — a shorter menu, but not a poor one, and Anthropic has said it will put more compute behind Claude in Cursor to cover the difference.

What we would take from it is the smaller, more durable point. Two weeks ago we wrote that the SpaceX acquisition’s real risks were the ones you couldn’t see from the outside, and that Cursor was busy making itself the place your code lives. This is that risk showing up on schedule: a deal you weren’t part of, removing a model you might have been relying on, with a date attached. It’s an argument for keeping your work portable — not for panicking about which of the Cursor models is best this week.

We’d also note the timing sits oddly against everything else happening this month, with Cursor, Replit and Ramp all building automatic model routers precisely so you don’t think about which model you’re using. The industry is working hard to make the model invisible, right up until one of them is taken away.

Not sure Cursor is the right tool for what you’re building? Take the 60-second Vibe Coding Tool Finder quiz →

Some links on this page are affiliate links. If you buy through them we may earn a commission at no extra cost to you.

FAQ