OpenAI is cutting native access to Cursor. Here's how developers can still use OpenAI, Claude, and other models via API keys and gateways.
What happened between OpenAI and Cursor? #
OpenAI announced it will wind down its contract that let Cursor offer OpenAI models natively inside the platform. The announcement landed on a Friday evening, a timing choice that usually signals a company trying to minimize attention. OpenAI’s stated reason: Cursor’s parent company, now owned by SpaceX (Elon Musk’s company, following its acquisition of Cursor), has a documented history of violating OpenAI’s terms of service, including a past distillation dispute involving xAI. OpenAI said it “cannot be confident” that SpaceX will use its technology within the bounds of its terms.
Cursor users aren’t locked out of OpenAI models entirely. What’s disappearing is the built-in, native integration where Cursor calls OpenAI on your behalf and bills you through its own subscription. Anyone who wants to keep using GPT-5.x or future OpenAI models inside Cursor can still do it, just not the way most people are used to doing it.
TL;DR #
OpenAI is severing its direct contract with Cursor, citing concerns that SpaceX (which acquired Cursor) could misuse OpenAI’s models given Musk’s companies’ history of breaking prior agreements, including a distillation dispute involving xAI.Native, in-app access to OpenAI models through Cursor’s own billing is going away, but developers can still reach OpenAI models by plugging in their own API key.** This isn’t an isolated incident**: Anthropic has already restricted Claude access for OpenAI, for xAI, and reportedly tightened access after Google acquired Windsurf, a similar AI coding IDE.The underlying fear is distillation, the practice of using a rival’s model outputs as training data to cheaply build a nearly-as-capable competitor, something Anthropic has publicly accused Chinese labs of doing at scale.Cursor’s CEO disputed the framing, noting OpenAI models represent about 5% of Cursor’s user traffic, though token volume and revenue share aren’t the same measurement and likely understate OpenAI’s actual dollar value to Cursor.The bigger trend is AI coding platforms verticalizing: model providers increasingly want their frontier models tied to their own tools, not powering a rival’s ecosystem, especially once that rival is owned by a direct competitor.Bring-your-own-key setups and model gateways become the practical workaround for developers who want to keep multi-model flexibility as native integrations get pulled one by one.
Remy is new. The platform isn't. #
Remy is the latest expression of years of platform work. Not a hastily wrapped LLM.
Why is OpenAI cutting off Cursor specifically? #
The immediate trigger is Cursor’s acquisition by SpaceX. Once Cursor sits inside Elon Musk’s corporate structure, alongside xAI, OpenAI treats it as a competitor with direct access to its models and, by extension, to the query data that flows through those models.
OpenAI’s blog post pointed to two specific incidents as justification. After Musk acquired Twitter (now part of SpaceX), OpenAI says the company broke terms of an existing contract. Separately, under oath, Musk reportedly admitted that xAI (also now under SpaceX) had violated OpenAI’s terms of service around distillation, the practice of querying a rival’s model heavily enough to extract training data for a smaller, cheaper model that approximates the original’s capabilities.
Distillation is a live industry fear right now, not a hypothetical one. Anthropic has publicly said it observed industrial-scale distillation attempts against Claude, describing large numbers of accounts systematically querying its models to harvest training pairs. Once you view Cursor as a pipeline that could route enormous volumes of coding queries into OpenAI’s models and back out to a Musk-owned lab, the calculus changes. OpenAI has spent billions training its frontier models. Handing a competitor a shortcut to replicate that capability, especially one it says has broken contract terms before, is a risk OpenAI decided wasn’t worth the revenue.
Is this actually unprecedented? #
No, and that’s part of what makes it notable. This is the latest in a pattern that’s played out repeatedly over the past year or so among AI coding platforms and model providers:
Google acquired Windsurf, another AI coding IDE that, like Cursor, didn’t build its own frontier models and instead depended on providers like Anthropic and OpenAI to power its product. Shortly after, Anthropic reportedly limited direct Claude access for Windsurf. Anthropic had already revoked OpenAI’s own direct access to Claude models, reasoning that OpenAI’s usage looked less like benchmarking and more like an attempt to study Claude’s outputs. And separately, Anthropic cut off xAI’s access to its models months before the Cursor story broke, for the same underlying reason: not wanting a competitor to use Claude outputs to close the capability gap.
Seen together, the pattern is consistent: as soon as a coding platform becomes tied to a company that competes directly with a model provider, that provider pulls native access. The Cursor and OpenAI situation is just the most visible recent example because of the personalities involved (Musk and Sam Altman have a long, public rivalry dating back to OpenAI’s 2015 founding, Musk’s departure in 2018, and a lawsuit Musk filed in 2024 over OpenAI’s shift to a for-profit structure).
How much does this actually affect Cursor users? #
- ✕a coding agent
- ✕no-code
- ✕vibe coding
- ✕a faster Cursor
The one that tells the coding agents what to build.
Cursor’s CEO, Michael Truel, pushed back on the framing, stating that OpenAI models account for roughly 5% of Cursor’s user traffic. That number sounds small, but it’s measuring token volume, not revenue or value delivered. Smaller and less capable models often need many more tokens to complete the same task, which inflates their share of raw traffic. A frontier OpenAI model doing a hard coding task efficiently in fewer tokens can represent a disproportionately large share of what users actually pay for, even if its token count looks modest next to cheaper models.
In practical terms, the users most affected are the ones who specifically pick an OpenAI model for certain tasks inside Cursor’s native model picker. That workflow disappears. The models themselves don’t become unreachable, they just require a different plumbing setup.
How do you keep using OpenAI (or any model) in Cursor now? #
The practical fix is the same one that’s existed for a while for anyone who wants provider flexibility inside an AI coding IDE: use your own API key instead of relying on the platform’s built-in, bundled access.
Bring your own API key. Most AI coding IDEs, Cursor included, support entering a personal API key for a given provider. You sign up for API access directly with OpenAI, Anthropic, Google, or another provider, generate a key, and paste it into the IDE’s settings. Usage bills straight to your account with that provider instead of running through the IDE’s subscription. This is exactly the workaround still available to Cursor users who want OpenAI models after the native integration ends.
Use a model gateway or router. Rather than juggling separate keys for every provider, a gateway service sits between your IDE and multiple model providers, letting you switch models without reconfiguring each one individually. This adds a layer of abstraction that keeps you provider-agnostic even as platforms narrow their native offerings.
Expect more of this, not less. As coding platforms get acquired by or aligned with specific model labs (Cursor with SpaceX/xAI, Windsurf with Google), the labs behind them have incentive to eventually push their own models as the default and deprioritize rivals’ native support. Relying on your own keys or a gateway is the way to insulate your workflow from decisions made at the corporate ownership level.
Is Cursor still worth using without native OpenAI access? #
For most workflows, yes. Cursor still supports Anthropic’s Claude models, Google’s Gemini models, its own in-house model, and, now that it’s under SpaceX, xAI’s Grok models, including Grok 4.6, which has become a genuinely competitive coding model even if it isn’t quite matching the very top of the frontier. Losing native OpenAI support is a real inconvenience for developers who specifically preferred GPT-5.x for certain tasks, but it’s a configuration problem, not a capability wall. Anyone willing to manage their own API key retains full access. The bigger question for the industry is whether this pattern accelerates. If every AI coding platform ends up owned by or tightly bound to a single model lab, the promise of one IDE giving you frontier-level choice across providers gets harder to sustain. That makes independent, provider-agnostic tooling and personal API key management more valuable, not less.
Frequently Asked Questions #
Can I still use GPT-5 or other OpenAI models in Cursor?
Yes, but not through Cursor’s native, built-in integration once the wind-down takes effect. You’ll need to supply your own OpenAI API key and pay OpenAI directly for usage rather than having it bundled into Cursor’s subscription.
Why did OpenAI cut off Cursor’s native access?
Seven tools to build an app. Or just Remy. #
Editor, preview, AI agents, deploy — all in one tab. Nothing to install.
OpenAI says it can’t be confident that SpaceX, which now owns Cursor, will follow its terms of service, citing a history of contract violations by Musk’s companies, including a distillation-related dispute involving xAI.
Is this only happening to Cursor?
No. Anthropic has restricted Claude access for OpenAI, for xAI, and reportedly for Windsurf after Google’s acquisition of that platform. It reflects a broader trend of model providers limiting access once a coding platform is owned by or aligned with a competing AI lab.
What is distillation, and why does it matter here?
Distillation is the practice of querying another company’s AI model heavily enough to collect input-output pairs, then using that data to train a cheaper model that approximates the original’s performance. Providers worry that giving a competitor’s platform native access effectively hands over free training data.
What’s the best way to keep multi-model flexibility in an AI IDE?
Use your own API keys for each provider you want, or route requests through a model gateway that supports multiple providers. This keeps your workflow independent of whatever ownership or partnership changes happen at the platform level.