Contract for OpenAI models in Cursor winds down

OpenAI is winding down the contract that supplies its models to Cursor, following the code editor's acquisition by SpaceX. The decision was set out in a post by OpenAI, which said it notified SpaceX of the wind-down on August 28, 2026.

The wind-down reads as a line drawn around ownership rather than around usage: a rival's models stop flowing into a product that a competing lab now controls. For Cursor, the practical exposure is limited by the fact that its stack has been moving toward Grok since the spring. For the wider market, it is a reminder that model supply agreements likely carry corporate-control conditions that surface only when an acquisition closes.

OpenAI describes the step as winding down the contract under which it provided its models to Cursor, and ties it directly to the change of ownership at the editor. In its post, OpenAI said the notification went to SpaceX on August 28, 2026.

SpaceX officially completed its acquisition of Cursor on August 14, 2026. Cursor had been working with SpaceXAI before that: according to Cursor, a partnership aimed at accelerating model training efforts started in April, several months before the deal closed in August.

Inside the editor, the team's work is now split across Grok Build, Grok Bot and the Grok API, according to Cursor. Grok 4.6 already ships in the product, which places models from the SpaceX side of the group in Cursor while the OpenAI contract is wound down.

Contracts of this kind give a third-party product commercial access to a lab's models, which the product calls through an API instead of running systems of its own. Cursor has drawn on models from more than one outside lab, and the OpenAI contract covered that supply.

When OpenAI access ends

OpenAI's post does not give a date on which Cursor's access to its models ends, and it does not set out the commercial terms of the contract or the scope of the models covered. The available material also gives no figure for the value of the SpaceX purchase, and no timetable for moving remaining OpenAI workloads inside the editor.