# OpenAI ends its Cursor partnership after the SpaceX acquisition > Source: > Published: 2026-08-30 01:36:14+00:00 ## The announcement OpenAI has confirmed it is **terminating its partnership with Cursor** following Cursor’s acquisition by SpaceX. In the company’s own words (posted August 29, 2026): under their proposal, **Cursor’s direct access to OpenAI models would end on November 12, 2026**. OpenAI frames the shutoff date as what the contract allows - the maximum notice it must give - and says it recognizes the people most affected are the developers who rely on OpenAI models inside Cursor, offering to “go above and beyond” to support them through the transition. A widely shared summary posted by Tesla Owners Silicon Valley adds detail that is not in OpenAI’s public statement: OpenAI notified SpaceX the same day, the November 12 cutoff is the maximum notice the contract permits, and **future OpenAI models - including Astra - will not be licensed to Cursor**. OpenAI has not disputed that summary. ## The stated reason The publicly cited grounds are breach of conduct by “Elon companies”: OpenAI points to **X’s behavior after the Twitter acquisition and what xAI stated under oath** as violations of OpenAI’s terms. In other words, the model licensing is being treated as a privilege tied to corporate conduct across a family of companies, not a bilateral commercial relationship with Cursor itself. ### The reply, and what it signals Elon Musk’s public reply was dismissive - “I couldn’t care less” - paired with the familiar claim that OpenAI “stole an open source nonprofit.” Whatever the legal history, the commercial reality moves faster than any courtroom: from November 12, one of the most-used AI coding editors loses direct access to one of its founding model partners. ## What this means for developers Cursor users on OpenAI models have about 90 days of runway - and the migration path is exactly the multi-provider setup this site keeps recommending: Cursor also serves Anthropic, Google, and open-weight models, so switching is a configuration change, not a replatform. The sharper lesson is about **single-vendor dependencies**: API access to a closed model is a contract, and contracts get re-evaluated when the company on either side of the deal is acquired. Your editor’s model menu is now subject to M&A risk you do not control. ## The local angle Every closed-model dependency now has a visible failure mode: a boardroom decision, an acquisition, a terms dispute - and anyone can turn off the switch. Your editor, your API bill, your whole workflow degrades in a quarter because two companies you have no stake in collided. That is precisely why you need local AI and open-weight models: **open weights cannot be un-partnered**. [GLM-5.3](/models/glm-5-3-flash) opened its weights this week, and any rig that runs a 70B-class model locally keeps answering on November 12 regardless of what gets announced on X. The [rig finder](/find) matches your workload to hardware, and the [model catalog](/models) shows which open weights make sense as your fallback.