What happened
On August 28, OpenAI told SpaceX it plans to cut off Cursor’s direct access to OpenAI’s models, with November 12, 2026 proposed as the date access ends, according to OpenAI’s announcement. The notice follows SpaceX’s $60 billion acquisition of Cursor, a deal announced in June and completed in August.
OpenAI says it “cannot be confident that SpaceX will use [its] technology within [its] terms of service.” It points to what it describes as a pattern from other Musk-run companies, including Twitter’s past contract breach and Elon Musk’s own admission, under oath, that xAI violated OpenAI’s terms.
OpenAI also plans to withhold future model releases from Cursor during this window, including an unreleased model referred to as Astra, according to DevOps.com’s reporting. Cursor CEO Michael Truell responded that OpenAI’s models account for only about 5% of the platform’s user traffic, calling the decision disappointing given the companies’ history. That is Truell’s reported traffic figure.
Why it matters
If you build inside Cursor and rely on an OpenAI model, whether that’s a GPT model or one of OpenAI’s reasoning models, which work through a problem in steps before answering, this deadline affects your setup directly, not just the two companies involved. Cursor still supports other model families: Anthropic’s Claude Fable, Opus, and Sonnet, Google’s Gemini 3.1 Pro and Gemini 3.7 Flash, and Grok models. Anthropic has also said it will add more compute capacity, the server power needed to run its models, for Claude inside Cursor during the transition, according to DevOps.com.
Whatever the dispute’s outcome, the cutoff shows how a model-vendor relationship can change a coding workflow. Most AI coding agents do not own the models they run. Cursor offers third-party models alongside its own Composer models. Access to a third-party model still depends on the vendor agreement. When a business relationship between the agent and a model vendor changes, whatever you built your workflow around can shift underneath you on a deadline you did not set.
Who should care
Anyone building daily inside Cursor with an OpenAI model selected, especially if you have written custom rules, saved prompts, or automation tuned to how a specific OpenAI model responds. It matters less if you already default to a Claude or Gemini model in Cursor, since your day-to-day access is not the part changing.
What builders should do next
November 12 is the proposed cutoff. Compare alternative models before then so a change in access does not force an unprepared switch. Pick one real task you already do often in Cursor with an OpenAI model, something with a clear right answer, like fixing a specific bug or adding one function you can test. Run it once against a Claude model and once against a Gemini model in Cursor’s model picker, using the same prompt each time.
Compare two things: whether the output is correct on the first try or needs follow-up corrections, and how long each response takes to come back. That tells you what actually changes for your workflow, instead of guessing from the model names alone.
OpenAI’s timeline may still shift before November 12. The announcement gives a proposed date, rather than confirming the final access terms. But the underlying lesson holds regardless of how this specific standoff ends: the model behind your coding agent is a business relationship, not a fixed feature. Test your alternatives now, while you still have the runway to do it on your own schedule.
End of article