OpenAI Just Cut Off Cursor — and It Barely Matters

After SpaceX bought Cursor, OpenAI is pulling its models from the most popular AI coding tool. The problem: most Cursor power users already switched to Claude.

OpenAI gave Cursor 30 days. In a post published this week, the company said it will “wind down” the contract that pipes GPT models into Cursor, the AI code editor now owned by SpaceX. The stated reason is clean corporate language about “alignment of interests.” The real reason has a name, and it’s Elon Musk.

This is a business decision dressed up as a principle. Musk is suing OpenAI. Musk runs xAI, which builds Grok, which competes directly with GPT. Musk’s SpaceX just bought the single most visible showcase for OpenAI’s coding models. Of course OpenAI pulled the plug. What’s funny is how little that plug was carrying.

The move that sounds bigger than it is

Here’s the part OpenAI’s post doesn’t mention: Cursor stopped being a GPT-first product a long time ago. Walk into any engineering team that lives in Cursor and ask what model they’re running. Overwhelmingly, the answer is Claude. Anthropic’s Sonnet has been the default agentic coding brain inside Cursor for well over a year, because it was simply better at multi-file edits, tool calls, and not hallucinating APIs that don’t exist.

The numbers back that up. On SWE-bench Verified — the benchmark that actually measures whether a model can resolve real GitHub issues — Claude Sonnet 4.5 lands around 77%, ahead of GPT-5’s roughly 75%. Two points sounds trivial. In practice, on a 30-step agentic task where every step compounds, it’s the difference between a pull request that merges and one you have to babysit. Cursor’s own usage telemetry has reflected this for a while. GPT was the fallback, not the workhorse.

So when OpenAI says it’s cutting Cursor off, the honest translation is: OpenAI is removing the option most Cursor users had already stopped clicking.

Who actually gets hurt

Some people do lose something. If you’re on a team that standardized on GPT-5 for its cheaper token pricing, or you leaned on OpenAI’s models for the o-series reasoning traces during debugging, your workflow breaks in 30 days. Cursor will migrate you, but defaults will shift and your carefully tuned prompts may behave differently against a new model.

The bigger loser is arguably OpenAI. Cursor is a distribution channel — millions of developers forming daily habits around whatever model sits behind the tab-complete. Handing that entire audience to Anthropic and Google out of spite is a strange way to win a coding-model war you’re already losing on the benchmark that matters. OpenAI is protecting a competitor’s showcase by abandoning its own.

What Cursor does next

Cursor’s answer is obvious and easy: lean harder into Anthropic, add more Gemini, and quietly ship whatever xAI puts in front of it. Expect Grok to appear in the Cursor model picker within the next few months — that’s the entire strategic logic of the acquisition. Musk gets a first-party channel to push Grok Code into developers’ hands, and a live feed of exactly how coding agents get used in the wild.

Watch three things over the next three to six months:

Grok Code inside Cursor as a default tier. SpaceX didn’t buy Cursor to keep paying Anthropic. A first-party Grok coding model, subsidized to undercut everyone, is coming.
Anthropic pressing its lead. Claude Opus 4.5 and the Sonnet line already own agentic coding. Expect Anthropic to court Cursor harder now that GPT is out — better rate limits, priority access, deeper Claude Code integration.
Google’s Gemini 3 making a real run. Gemini’s giant context window is genuinely useful for whole-repo reasoning, and Google has the pricing muscle to buy market share. It’s the dark horse in Cursor’s picker.

What to do today if GPT was your default

You don’t have to wait for the wind-down. Switch your primary model in Cursor now and get ahead of the migration. In Cursor: open Settings → Models, disable the OpenAI models, and set Claude Sonnet 4.5 as your default agent model. If you want to route through your own key for higher limits, drop it in the API keys panel.

Want to sanity-check Claude on a real task before you commit? Hit the Anthropic API directly:

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-4-5",
    "max_tokens": 1024,
    "messages": [
      {"role": "user", "content": "Refactor this function to remove the nested loop and explain the time complexity change:\n\ndef find_dupes(a):\n    out = []\n    for i in a:\n        for j in a:\n            if i == j and a.index(i) != a.index(j):\n                out.append(i)\n    return out"}
    ]
  }'

If you’d rather keep every model behind one key so you’re never hostage to a single vendor’s mood, route Cursor through OpenRouter and set a fallback chain: Claude first, Gemini second. Then it doesn’t matter who cuts off whom next quarter.

The real story here

This isn’t about coding models. It’s about what happens when the AI industry’s infrastructure gets tangled in personal feuds. OpenAI and Musk have been at war since the lawsuit, and now that war is reaching down into the tools developers use to ship code on a Tuesday afternoon. Your editor’s model picker is now a proxy battlefield.

The overhyped read is that this cripples Cursor. It doesn’t. The under-discussed read is that OpenAI just voluntarily surrendered its biggest developer showcase to Anthropic and Musk in the same move — and did it while trailing on the one benchmark developers actually care about.

If you build for a living, the lesson is simpler than the drama: never let one company’s API be a single point of failure in your workflow. Wire up a fallback today. The models are close enough that switching costs you almost nothing — and, as this week proved, the vendor can walk away with 30 days’ notice over a fight that has nothing to do with you.