Cursor Loses OpenAI Models on November 12

OpenAI is ending the contract that gives Cursor direct access to its models, with the cutoff set for November 12, 2026. The decision follows SpaceX’s acquisition of the coding company, and OpenAI frames it as a question of trust in Elon Musk’s businesses.

Key Takeaways

• The cutoff lands on November 12, 2026 and covers Cursor’s direct access to OpenAI models.
• Those models account for roughly 5% of the tool’s AI traffic, which keeps the technical fallout limited.
• Anthropic announced within hours that it was raising compute to support Claude inside Cursor.

Five percent of the traffic, all of the signal

The break has a date on it. OpenAI stops supplying its models to Cursor on November 12, 2026, a few months after SpaceX bought the company for sixty billion dollars.

The reasoning is spelled out in the decision note OpenAI published: the company says it cannot be confident SpaceX will use its technology within the terms of service, drawing on its own history with Musk-owned businesses.

The two and a half month runway is a choice in itself. OpenAI could have cut access on short notice, and setting the date in November hands Cursor a full quarter to move its defaults, which reads less like a sanction than like a supplier closing an account in order.

The technical damage is modest. OpenAI models make up around 5% of Cursor’s AI traffic, and the tool has long run on a multi-vendor catalogue, an architecture that already let it absorb the load through GitHub’s six-hour outage.

Even so, 5% of the traffic on what has become the default assisted coding editor is a large number of calls. The loss counts less in immediate market share than in visibility among developers who meet a model for the first time inside the editor they open every morning.

That figure is a floor rather than a ceiling on the disruption. It measures the calls routed today, not the share of users who set an OpenAI model as their default and will have to make that choice again before the deadline.

Thibault Sottiaux, on the OpenAI side, put the position in one line: it boils down to trust. The wording targets Musk and his companies rather than the wider coding-tool ecosystem.

Nothing in the decision touches the rest of the catalogue. Cursor keeps its access to competing providers and to the open-weight models it hosts itself, so the product does not lose a capability, it loses one supplier among several.

Cursor’s position is also sturdier than the headline suggests. An editor built on a multi-vendor catalogue absorbs the loss of one provider by design, and that architecture was a product decision long before it became a defensive one.

The Twitter precedent, stated on the record

OpenAI does not stop at an abstract risk, it names two precedents. The first goes back to December 2022, when the Twitter acquisition was followed by the cutoff of OpenAI’s access to platform data, a contract worth around two million dollars a year.

The second grievance concerns Grok being trained on outputs produced by other labs’ models, against those labs’ terms of service. Trust was already the load-bearing issue when it sat at the centre of the Elon Musk versus OpenAI trial.

Both grievances share one logic. Neither concerns the partner’s technical quality, both concern how it behaved contractually once the balance of power shifted, which explains why the decision follows an acquisition rather than a usage incident.

Neither precedent involves a court. Both are commercial disputes OpenAI is choosing to settle by withdrawing supply rather than by litigating, which moves faster and leaves no ruling behind to constrain the next decision it takes.

Michael Truell, a Cursor cofounder and now an executive at SpaceX, answered on X. He said the company had counted on OpenAI’s platform staying neutral infrastructure and that talks with the OpenAI team were under way.

Truell is in an awkward spot. He is answering as the cofounder of the tool losing access while sitting on the executive team of the company whose acquisition triggered the loss, which leaves him defending a neutrality he no longer fully controls.

Musk dispatched the whole thing the next morning with a post saying he could not care less. The tone sits oddly against the May verdict that cleared OpenAI’s path to an IPO, and extends a rivalry the courts failed to cool.

The gap between the two responses says something on its own. OpenAI wrote a dated, documented decision note, and the owner of the acquiring company answered in a single line on a social platform, which sets the tone for whatever negotiation Truell says is under way.

The date is the only firm commitment on the table. Everything else, including whether those talks lead anywhere, rests with two companies whose leaders have spent the year answering each other in public rather than across a table.

More articles on Horizon

• Claude Sonnet 5 Fixes Claude Better Than 28 Researchers
• GLM-5.3-Flash Runs on 100,000 Chinese Chips
• Codex Prepares a Mode That Never Stops Working

Anthropic raises compute within hours

The fastest response did not come from Cursor. Within hours of the announcement, Anthropic said it was increasing the compute allocated to support Claude models inside the tool.

The arithmetic is easy to follow. Anthropic competes with both camps yet keeps a live partnership with SpaceX, which lets it pick up share a rival just walked away from for nothing in return. That closeness was already readable when the SpaceX AI phone prototype leaked and Musk denied it.

The move costs Anthropic very little and buys visibility. Raising compute for an editor that just lost a supplier is the cheapest available display of reliability, and it lands in front of exactly the developers now picking their default model.

For teams building on Cursor, the migration starts now. Two and a half months separate the announcement from the cutoff, plenty of room to shift default configurations onto Claude or an open model, not enough to leave the question alone until November.

Switching is not free either. Teams that tuned their prompts and project rules against one specific model will have to redo that calibration on another, a cost that never appears on an invoice and gets paid in hours.

The timing lands awkwardly for procurement as well. November sits at the end of the budget year for a lot of engineering teams, so the cutoff forces a supplier review at exactly the moment those budgets get locked for the year ahead.

For anyone selling model access, the episode raises a pricing question as much as a legal one. If a provider can withdraw on ownership grounds, buyers will value multi-vendor architectures more highly, and the premium a single frontier lab can charge for exclusivity falls accordingly.

The one group with nothing to gain here is the developer who picked Cursor precisely because it stayed above the vendor fight. That neutrality was the product promise, and it just turned out to depend on decisions made two levels above the user.

The substance runs deeper than the feud. OpenAI is pushing its own coding agent, pitched from launch as a tool built for enterprise desk work, and cutting off a distributor now owned by a rival protects that ground as much as it punishes a partner.

What remains is the precedent set for everyone else. A model provider has just shown it will pull access based on who owns the customer, which turns the neutrality of these platforms into a variable that moves with the next acquisition.

Follow the story on Horizon.