OpenAI is winding down Cursor. Here's what that means for AI coding and code review

OpenAI is winding down Cursor. Here's what that means for AI coding and code review

The shutoff date is November 12. Astra never arrives. The review layer should not be hostage to the same contract.

On August 28, OpenAI told SpaceX it will wind down the contract that supplies OpenAI models to Cursor. The proposed shutoff is November 12, 2026 — the maximum notice in the agreement. OpenAI's upcoming Astra model will not go to Cursor at all.

That is not a product tweak. It is the first hard answer to a question WIRED asked as soon as the deal was public: can Cursor stay a multi-model platform inside SpaceX? OpenAI just said no — at least for GPT.

Cursor users keep current OpenAI models for about ten more weeks. After that, GPT inside Cursor is gone, and the next OpenAI frontier model never shows up there. Coding agents and pull request review are not the same job. If you treat them as one vendor relationship, November 12 becomes a worse day than it needs to be.

One path is already uncoupled. ThinkReview MCP reviews you run from Cursor are unaffected. The agent in the editor may lose GPT. The review still runs on ThinkReview's models, not Cursor's OpenAI contract.

What OpenAI actually decided

OpenAI's post is short. The facts that matter:

• Current models stay until November 12. OpenAI says it is using the longest notice the contract allows so developers are not cut off overnight.
• Astra does not ship to Cursor. Future OpenAI models are out of the remaining window on purpose.
• The lever is change of control. OpenAI says its custom Cursor agreement included a limited window to cancel after an ownership change. SpaceX closed the $60 billion acquisition of Anysphere on August 14. Two weeks later, OpenAI pulled that clause.
• The stated reason is terms of service, not taste. OpenAI writes it cannot be confident SpaceX will use its technology within its terms, citing Musk companies' history: Twitter/X breaking contract terms after the acquisition, and Musk admitting under oath that xAI — now also inside SpaceX — distilled OpenAI data in violation of those terms.

OpenAI also says the decision was tough. Cursor went through OpenAI's startup accelerator in 2024. The companies worked together for nearly four years. The people who feel this first are the developers who still pick GPT inside Cursor every day — not the labs arguing about distillation.

Respect for the Cursor team is in the post. Access for Astra is not. Those two sentences can both be true.

Why this was always going to happen

SpaceX did not buy a text editor. It bought distribution: millions of developers, a coding agent that already sits in the loop, and a place to run Grok next to the work. Cursor joining SpaceXAI is meant to feed Grok Build, Grok Bot, the Grok API, and Cursor itself. Every token Cursor still sends to OpenAI or Anthropic is revenue and training signal leaving that stack.

That incentive was obvious on close day. Coverage at the time noted the threat to multi-model support could come from either side: SpaceX deprioritizing rival APIs, or OpenAI and Anthropic refusing to subsidize a direct competitor. OpenAI moved first. Anthropic has not made a matching public cut — yet. Developers should not wait for a second letter to assume the platform stays "model-neutral."

OpenAI's Astra argument is the other half. Frontier models are now treated as assets that need a trusted operator, not just an API key. A change-of-control clause is how a lab keeps a captured IDE from becoming the default channel for the next model. Whether you buy OpenAI's ToS story or see it as competitive fencing, the operational result is the same: Cursor is no longer a reliable GPT client after mid-November.

Grok does not need that client. Grok 4.6 already runs in Cursor, and SpaceXAI has every reason to make it the default. That is a strong coding model. It is not a reason to let one lab own both the editor and the review.

What developers should change before November 12

Do not panic-migrate the IDE on Monday. Do inventory the places GPT is load-bearing.

• Separate coding from review. If Cursor is where you write, it does not have to be where you judge the PR. An IDE cutoff should not also kill your review model.
• Keep using ThinkReview MCP in Cursor. Asking Cursor to call review_url_code on a PR still hits ThinkReview, not OpenAI-via-Cursor. That workflow does not shut off on November 12.
• Keep a GPT path that is not Cursor's model picker. OpenAI still wants developers on its models — just not through SpaceX. ChatGPT, the API, Codex, ThinkReview, and tools that call OpenAI directly are a different contract.
• Assume Astra never lands in Cursor. Plan the next frontier GPT cycle as if Cursor will not have it. If your team was waiting on Astra-in-Cursor, that wait is over.
• Watch Anthropic the same way. Claude in Cursor is still up. The incentive geometry did not change. Have a non-Cursor Claude path before you need one.
• Treat Grok as a first-class option, not a consolation prize. For a lot of agentic coding, Grok 4.6 is already the model people reach for. Use it where it is strong. Do not use the OpenAI news as a reason to stop comparing.

The failure mode is coupling. Teams that bound "how we write code" and "how we review code" to one vendor's Cursor integration will spend November rewriting both.

What this means for code review

Pull request review is the wrong place to inherit an IDE vendor fight.

A coding agent in Cursor is trying to finish a task. A reviewer is trying to catch the task that should not merge. Those jobs want different models on different days: GPT for one repo, Grok 4.6 for another, Claude or Gemini when the diff is messy. That only works if model selection lives outside the captured client.

That is the job ThinkReview is built for: a copilot on the PR page — GitHub, GitLab, Azure DevOps, and Bitbucket — not a bot that rubber-stamps the merge, and not a sidecar that dies when an IDE contract does. The same split holds if you never leave Cursor: ThinkReview MCP is a client in the editor. The review itself is ThinkReview fetching the diff and running the model you picked in the portal.

• MCP reviews in Cursor are unaffected. OpenAI is ending GPT as Cursor's coding model. It is not ending ThinkReview. Paste a PR URL, call review_url_code, get the same structured review — GPT-5.6 included, if that is what you selected. November 12 does not change that.
• Pick the review model independently. GPT-5.6, Grok 4.6, Claude, Gemini, Kimi, and the rest stay in Model selection. OpenAI leaving Cursor does not remove OpenAI from ThinkReview.
• Keep Cursor for implementation. Cursor deeplinks still turn a finding into a ready prompt in the editor. Review with one lab, implement with another. That split just got more valuable.
• Do not wait for Astra to show up in the IDE. If you want the next OpenAI model on a real diff, select it where the PR lives — not where the agent writes the first draft.
• Keep a human on the loop. Vendor drama does not change the review bar. Treat severity labels as input. Pair any model with repository-level context when the real bug lives outside the hunk.

OpenAI gave Cursor the longest notice the contract allowed. Use the weeks. Diversify the review path now, keep the editor you actually like, and stop treating "GPT in Cursor" as the same thing as "GPT on the pull request." If you already review PRs from Cursor through ThinkReview MCP, that path does not need a migration.

Pick the model in Model selection. Run it on a real diff. Keep the judgment.

Details reference OpenAI's announcement. Ready to review the next PR with a model that is not tied to one IDE contract? Install ThinkReview or connect ThinkReview MCP in Cursor.