OpenAI drops out of Cursor: what really changes for developers

I'm working on a project, I open the model selector and switch between GPT and Claude depending on what I have in front of me. A heavy query, a refactor, documentation. It became completely normal.

That's why the OpenAI and Cursor news caught my attention. Not because the editor is going to stop working tomorrow. The interesting part is underneath: we're building development workflows on top of business agreements that can disappear even though technically everything works perfectly.

What OpenAI actually announced

On August 28, OpenAI announced that it intends to terminate the contract through which it provides its models directly to Cursor.

The proposed cutoff date is November 12, 2026.

Until then, OpenAI says it will continue providing Cursor access to the models it currently uses, although Cursor could decide to end it earlier. It also clarified something important: OpenAI's future models are outside the agreement.

That last part may end up being more relevant than losing a specific model.

A GPT that works well today will keep aging. If in a few months a clearly superior model comes out and Cursor can't offer it through its usual integration, the difference will start to be felt.

SpaceX enters the story

The move happened after SpaceX's acquisition of Cursor.

OpenAI had a clause in its agreement that allowed it to cancel the contract during a limited period after a change of control. It decided to use it.

The public explanation was quite undiplomatic.

OpenAI says that, due to past experiences with Elon Musk's companies, it doesn't have enough confidence that its technology will be used in compliance with its terms of service.

Here we're completely outside the technical terrain.

Cursor may have an excellent architecture. GPT may work perfectly inside the editor. Neither of those two things resolves a contractual conflict between the companies behind them.

That's precisely the point worth observing.

Cursor depends less on OpenAI than it seems

There's another interesting fact.

According to Cursor, OpenAI's models account for about 5% of total model traffic within its platform.

That changes the reading quite a bit.

Cursor has been turning into a layer where the developer chooses the model they need. Anthropic, Google, and other providers are part of that strategy, in addition to Cursor's own models.

So losing the direct relationship with OpenAI hurts, but it hardly leaves the editor unusable.

The situation also shows where these tools are evolving: the editor matters, but the model is becoming an interchangeable piece.

At least it should be.

GPT could still be used inside Cursor

There were quite a few headlines saying that OpenAI "removed GPT from Cursor."

Not yet.

OpenAI even published documentation explaining how to keep using its models inside the editor.

There are three main paths:

• Use your own OpenAI API key.
• Install the Codex extension inside Cursor.
• Connect through a compatible provider that acts as a gateway.

An API is, explained without getting too deep into programming, a controlled door so two systems can talk to each other.

The difference is who opens that door.

Until now Cursor could directly manage that relationship with OpenAI. With your own API, the relationship becomes between the developer and OpenAI.

That also means different billing and terms.

The real problem is dependency

I've been seeing this for years with technologies far less flashy than artificial intelligence.

A library stops being maintained. A provider changes prices. A platform removes a feature. Microsoft changes an API. A service that seemed permanent ends up discontinued.

With AI we're adding another dependency to the stack, except it changes much faster.

Not long ago the discussion was which model generated better code. Now we also have to think about who controls access to that model.

My preference is quite clear: I want to be able to swap out any of those pieces without rebuilding my entire workflow.

This also changes how to choose AI tools

For a while we evaluated AI editors almost the way we evaluated Visual Studio versus VS Code.

Features, extensions, performance, price.

Now I'd add something else to the list: how easy it is to leave.

It sounds strange to choose a tool thinking about how to abandon it, but in software we do exactly that when we design an integration properly.

If my application depends on an external provider, I try to place a layer between my system and that provider. If it changes tomorrow, I replace that part.

With AI tools we should start thinking along the same lines.

An editor that supports Claude, GPT, Gemini, custom models, and external connections has an advantage that doesn't easily show up in a benchmark: it lowers the cost of switching.

What I take away

This story can still change. Cursor has said it's in talks with OpenAI to try to resolve the situation, and the November 12 date is still on the table as a proposed date.

But even if they reach another agreement, the notice has already left a pretty useful lesson.

The model we use every day today can disappear from a tool for reasons that have absolutely nothing to do with programming.

Contract. Acquisition. Competition. Business strategy.

The code stays exactly the same.

Our workflow, maybe not.

Designing so you can also switch

I don't plan to stop using a tool just because it depends on external services. That would be unrealistic.

But I do try to keep a single combination from becoming indispensable.

If tomorrow GPT is no longer directly integrated into Cursor, there are alternatives. If Claude changes its terms, there should be another way out. And if the editor itself stops fitting my work, my projects should stay with me.

The battles between companies will continue.

I'd rather my code not have to take part in them.