OpenAI Plans to End Cursor’s Direct Model Access After SpaceX Acquisition

1. OpenAI has proposed a November 12 cutoff

OpenAI plans to terminate the contract that gives Cursor direct access to its models following the coding company’s acquisition by SpaceX. The proposed shutoff date is November 12, 2026, but that date is not yet final.

OpenAI announced the decision on August 28 and said it had notified SpaceX that day. Its subsequent support guidance clarified that OpenAI intends to keep supplying the models Cursor already uses during the transition. Cursor could end that access earlier, and OpenAI said it would publish the official termination date once both companies confirm it.

The distinction matters. OpenAI is not immediately disabling every way to use its models inside the Cursor application. It is winding down the commercial agreement under which Cursor supplies and routes those models as part of its own service.

OpenAI also said Cursor will not receive future OpenAI models under the agreement. At publication, Cursor’s documentation lists GPT-5.6 Luna, GPT-5.6 Sol and GPT-5.6 Terra among its available models. Sol has a listed maximum context of one million tokens, while all three have default contexts of 272,000 tokens.

The planned termination therefore creates two separate changes: direct access to the current model set would end after the transition, and new OpenAI releases would no longer enter Cursor through the existing partnership.

2. The SpaceX acquisition activated a contractual exit window

Cursor confirmed on August 14 that SpaceX had completed its acquisition of the company. The transaction followed an April partnership focused on using SpaceXAI infrastructure to expand Cursor’s model-training capacity.

The acquisition was valued at $60 billion in reporting based on SpaceX’s regulatory disclosures. Cursor said the combination would give it access to what it described as the world’s largest fleet of GPUs, enabling it to train stronger models and operate them more economically. It identified Grok 4.6 as an early product of the companies’ work together.

OpenAI said its custom Cursor agreement contains a limited cancellation window following a change of control. It is invoking that provision while giving what it described as the maximum notice permitted by the contract. The companies had worked together for nearly four years.

OpenAI framed the decision as a contractual and safety issue. It said it could not be confident that SpaceX would use OpenAI technology within its terms of service, citing what it characterized as earlier contract violations by Musk-controlled companies. Those are OpenAI’s allegations and rationale; its announcement did not disclose the Cursor contract or identify a breach by Cursor under SpaceX’s ownership.

The decision is also tied to Astra, an unreleased OpenAI model. On August 7, OpenAI said preliminary internal evaluations showed that Astra’s agentic coding and cybersecurity performance was strong enough that the company could not rule out “Critical” cyber capabilities under its Preparedness Framework. OpenAI subsequently cited its responsibility to control Astra’s use when explaining why future models would not be supplied to Cursor.

Cursor co-founder Michael Truell said OpenAI models account for about 5% of Cursor user traffic and that the companies were discussing a possible resolution. The company has not defined whether that percentage represents requests, tokens, users or spending, so it should not be treated as a complete measure of customer dependence.

3. Cursor remains a multi-model product

The loss of OpenAI’s direct supply would not remove Cursor’s own models or its integrations with other providers. Cursor’s current catalog includes Claude models from Anthropic, Gemini models from Google, its Composer family and Grok models developed within the SpaceXAI organization.

That diversity limits the effect on Cursor’s total request volume, particularly if Truell’s 5% figure accurately describes current usage. It does not make the OpenAI models interchangeable with their replacements. Coding teams may depend on a particular model’s reasoning behavior, context limits, tool use, latency or compatibility with existing prompts and rules.

The change also affects Cursor’s position as a model-neutral interface. Before the acquisition, developers could select models supplied by competing AI laboratories within one product and one billing relationship. After the proposed cutoff, OpenAI models would require a separate access and payment route, while Cursor’s own service would remain responsible for the other models it supplies.

For enterprise customers, aggregate traffic is less informative than organization-level usage. A company could have little exposure to OpenAI across its entire Cursor account while still relying on GPT models for a particular repository, automated workflow or team. Those dependencies need to be identified before direct access changes.

4. Developers can still use OpenAI models through narrower routes

OpenAI has documented three ways to retain some OpenAI access inside Cursor: supplying an OpenAI API key, installing the Codex IDE extension or connecting through a compatible AI gateway.

A user-supplied API key allows supported OpenAI models to run in Cursor’s local Chat and Agent features. Requests are billed to the user’s OpenAI API account rather than included in a Cursor or ChatGPT subscription. Available models and limits depend on the user’s OpenAI API organization.

This is not a complete substitute for Cursor-supplied access. OpenAI says personal API keys do not apply to Cursor Tab or autocomplete, Auto routing, Cloud or Background Agents, Automations, the Cursor CLI, or Cursor’s API and SDK. Cursor’s own documentation further describes OpenAI key support as covering standard, non-reasoning chat models rather than every OpenAI model.

The data and billing relationship also changes. Cursor says requests made with a personal key still pass through its backend for final prompt construction, although the key is encrypted and not retained after the request. Cursor’s Zero Data Retention commitment does not apply to those requests; data handling instead follows the selected model provider’s policy.

The Codex IDE extension offers a separate route. It can run inside Cursor using an eligible ChatGPT subscription or an OpenAI API account, but it provides its own agent interface. Installing it does not restore OpenAI models to Cursor Chat, Agent, Tab, Auto or cloud services.

Organizations can also configure Azure, Amazon Bedrock or another compatible gateway. That option can preserve centralized credentials and provider billing, but it remains subject to Cursor’s local-feature limitations and to the models exposed by the gateway.

5. The practical impact depends on the workflow, not the installation

Developers who use Cursor primarily with Claude, Gemini, Composer or Grok may see little immediate disruption. Users who select GPT models through Cursor’s model picker will need to change models, adopt separate API billing, use the Codex extension or move the affected work to another OpenAI-supported environment.

Cloud-dependent workflows face the clearest limitation because the documented alternatives cover local Chat and Agent use rather than Cursor’s cloud agents, automations, CLI or programmatic interfaces. A personal API key cannot preserve the same OpenAI-backed behavior in those features.

The proposal also freezes Cursor’s direct OpenAI access at the current model generation during the transition. Even if the November date moves through negotiations, OpenAI’s announced position is that future models—including Astra—will not be supplied under the existing agreement.

For companies, the relevant migration unit is therefore each model-dependent workflow rather than each Cursor seat. Teams using OpenAI models need to distinguish local interactive work, which has documented alternatives, from routed and cloud functionality that would require a different model or development platform.

Frequently Asked Questions

When will OpenAI models stop working directly in Cursor?

OpenAI has proposed November 12, 2026. The date is not final, and Cursor could end direct access earlier.

Will every OpenAI feature inside Cursor disappear?

No. Users can retain limited access through an OpenAI API key, the Codex IDE extension or a compatible gateway. These routes do not restore OpenAI models across all Cursor features.

Can a ChatGPT subscription pay for OpenAI usage in Cursor Chat?

No. Cursor Chat requests made with an OpenAI API key are billed separately through the API account. An eligible ChatGPT subscription can instead authenticate the separate Codex IDE extension.

Which Cursor features cannot use a personal OpenAI API key?

Cursor Tab, Auto routing, Cloud and Background Agents, Automations, the CLI, and Cursor’s API and SDK cannot use the key.

Is Cursor negotiating with OpenAI?

Yes. Cursor co-founder Michael Truell said the companies were discussing a resolution, while OpenAI said it would publish the final termination date once confirmed.

Sources

• Original OpenAI announcement on X
• Our decision on Cursor following its acquisition by SpaceX — OpenAI
• Using OpenAI models in Cursor — OpenAI Help Center
• Responding to the next frontier of critical cyber capabilities — OpenAI
• Cursor is now a part of SpaceX — Cursor
• Cursor Documentation and current model catalog
• Bring your own API key — Cursor Docs
• SpaceX buys AI coding startup Cursor for $60 billion — Associated Press
• OpenAI Pulls Its AI Models From SpaceX-Owned Cursor — The Information