OpenAI Cuts Off Cursor, Anthropic Steps In

OpenAI is walking away from Cursor. Two weeks after SpaceX closed its $60 billion acquisition of Anysphere, Cursor's parent company, OpenAI told SpaceX it's winding down the contract that supplies GPT models inside the editor. Every future model is locked out too, including the newly launched GPT-6 Astra, permanently. Cursor says the immediate damage is small: OpenAI models carry about 5% of its traffic. But the dispute has pulled Anthropic into the middle of it, using the moment to push more Claude compute into Cursor, some of it running on hardware SpaceX itself operates.

What changed

SpaceX closed its all-stock acquisition of Anysphere on August 15, 2026, the largest acquisition of a venture-backed startup on record, and folded Cursor into a new SpaceXAI division. The deal triggered a change-of-control clause in Cursor's contract with OpenAI. Two weeks later, on August 28, OpenAI notified SpaceX it was terminating the model-supply agreement, giving the maximum notice period the contract allows. Cursor's direct access to current OpenAI models is set to end November 12, 2026, and OpenAI says no future models, including GPT-6 Astra, will be supplied to Cursor at all.

OpenAI's stated reason wasn't technical. It was trust: "We cannot be confident that SpaceX will use our technology within the terms of our service agreement, a judgment informed by our history of contract breaches with companies owned by Musk." OpenAI pointed to a Twitter data-licensing deal Musk cancelled in 2022 and to Musk's own admission, during cross-examination in his lawsuit against OpenAI, that xAI trained on OpenAI's outputs.

openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/

Why it matters

OpenAI is giving up a relationship it estimated was worth more than $1 billion a year in revenue to make a point about who it will and won't do business with. That's a new kind of leverage in the AI coding-tool market: model access as a contractual weapon, pulled the moment a competitor's owner takes a stake in a distribution channel. It's also a preview of what happens once a model crosses into genuinely sensitive territory. GPT-6 Astra is the first model to hit OpenAI's "Critical" capability threshold for cybersecurity under its Preparedness Framework. During internal testing, it found and used two previously unknown zero-day vulnerabilities on its own. OpenAI is now the one deciding which companies get anywhere near a model like that, and it just decided SpaceX doesn't.

The 5% traffic number cuts against the drama, though. Cursor was already a multi-model router before any of this, with Claude, Gemini, and its own Composer models carrying most of the load. Losing OpenAI is a real loss of choice for developers who specifically wanted GPT models in that editor, but it isn't an outage. What it does change is who Cursor answers to going forward.

Anthropic confirmed it will keep increasing compute for Claude inside Cursor. The same week, it separately doubled Claude Code's five-hour rate limits for Pro, Max, Team, and Enterprise accounts, funded in part by a compute arrangement that runs through Colossus 1, the 220,000-plus GPU cluster SpaceX operates. The company that triggered OpenAI's exit is now, indirectly, helping power the tool replacing it. That's the kind of tangle a press release doesn't spell out.

What people are saying

Cursor co-founder Michael Truell downplayed the practical impact while confirming talks are ongoing:

"OpenAI models serve about 5% of Cursor user traffic, and we're speaking with the OpenAI team to resolve this."

Anthropic's Tom Brown used the opening to reaffirm the Claude relationship:

"We'll continue to increase compute to support Claude models in Cursor and are excited for what comes next with them at SpaceX."

Developer Michael Guo pointed out the timing implies something about Astra itself:

"This almost suggests OpenAI's next frontier model, Astra, will arrive before November 12."

What it looks like in practice

Open Cursor's model picker today and GPT-6 Astra, GPT-5.6 Sol, and older OpenAI options sit next to Claude Fable 5.1, Gemini, Grok, and Composer. After November 12, that picker loses the OpenAI column entirely, and it stays gone for every model OpenAI ships from now on, not just the current lineup.

For a team that built prompts and evals around a specific GPT model inside Cursor, that's real migration work: pulling up every saved prompt, custom instruction, and eval set written against a GPT model, then re-running it against Claude or Gemini to see what breaks. A prompt tuned around one model's particular phrasing habits doesn't always get the same result from another, so "swap the model" rarely means the work is done in one step. For a team already running most of its work through Composer or Claude in Cursor, the change shows up as nothing more than one greyed-out option in a dropdown.

What to do about it

• If your team routes specific tasks to GPT models inside Cursor, start re-testing those prompts against Claude or Gemini now rather than in November.
• If you specifically want GPT-6 Astra for coding work, plan to reach it through the OpenAI API, ChatGPT, or a tool like Codex directly, since it won't arrive in Cursor at all.
• Reassess how much your team's workflow depends on one editor's model roster. Cursor's pitch has always been router flexibility; this is a live test of what that's worth when one provider walks away.
• Watch for a resolution before November 12. Truell says Cursor and OpenAI are still talking, and the deadline is a contractual maximum, not a fixed outcome.

The short version

OpenAI is cutting off Cursor's access to its models, current and future, because SpaceX now owns it. Cursor says the practical hit is small since OpenAI was only 5% of its traffic, but Anthropic is already filling the gap with more Claude compute, some of it running on SpaceX's own hardware.