How do you connect GPT to Cursor when model access is changing? In a compatible bring-your-own-key mode, you can configure Kumo as the provider: paste a key, replace the OpenAI Base URL and choose a supported GPT model. We explain OpenAI's announcement, how to verify a connection today and why a working API matters more than promises about what happens after November.
GPT is leaving Cursor: what actually happened?
On August 28, 2026, OpenAI announced its intention to end its direct agreement with Cursor following the editor's acquisition by SpaceX. The proposed cutoff is November 12, 2026. The company linked its decision to concerns about Elon Musk's companies respecting its terms. This is an OpenAI statement, not a personal quote from Sam Altman.
For developers, the practical question is how to keep using GPT in a familiar project if the built-in model selection changes. Start by checking your connection method. Choosing a model from the built-in list and making a request through your own API key are different access mechanisms.
Three things that are easy to confuse
| What you have | What it means | Relationship to Kumo |
|---|---|---|
| Cursor subscription | Access to editor features and included model usage | Kumo does not change this subscription's terms |
| ChatGPT subscription | Access to the ChatGPT product under its terms | It is not a Kumo key or balance |
| Kumo API key | Access to supported models with token billing | Used in a compatible own-key configuration |
Changing how you pay for a model does not require moving your repository. That is the point of API access: choose the tool you enjoy coding in separately from the service that bills compatible requests. The editor must still support the selected model and protocol.
How to connect GPT to Cursor through Kumo
- Create an account in the Kumo dashboard, fund your balance and issue a separate API key for Cursor.
- Open Kumo's catalogue and copy the exact GPT model identifier. Confirm that it supports access through your own OpenAI-compatible key in your Cursor version.
- Go to Cursor Settings → Models → API Keys or the Providers section. Paste your Kumo key into OpenAI API Key and enable it.
- Enable Override OpenAI Base URL and enter https://api.kumorouter.com/v1. Do not use the dashboard or Kumo homepage address.
- Add the copied identifier through Add custom model if it is not already listed. Save the settings and run Verify when that check is available.
- Open a new chat, explicitly select the connected GPT model and send a short request. For example, ask it to explain a function's accepted inputs and return value.
- Open Kumo's request history. Check the model, call time and charge: this confirms that the request used your balance.
Which fields should you change?
| Cursor field | What to enter | Common mistake |
|---|---|---|
| OpenAI API Key | Your complete Kumo API key | Entering an account password or another service's key |
| Override OpenAI Base URL | https://api.kumorouter.com/v1 | Omitting /v1 or appending an endpoint path |
| Add custom model | The exact technical identifier from the catalogue | Entering a display name instead of an identifier |
| Model in chat | Your connected GPT model | Leaving Auto or another built-in model selected |
Setup details are in Kumo's Cursor documentation. Do not guess a model identifier from a name in the news. The available API catalogue and the models accepted by a particular client can differ.
Why an API key alone is sometimes insufficient
Cursor restricts how custom keys can be used. Its official documentation describes chat model support and specifically lists standard non-reasoning models for OpenAI. Tab uses the editor's built-in models.
That means you cannot assume every GPT model in Kumo's catalogue will automatically work in every Cursor mode. For models with special capabilities, check your exact workflow: ordinary chat, tool calling and agent work can have different compatibility.
What if your required GPT model will not connect?
Start with the basics: is the key correct, does the address end with /v1 and have you selected the exact identifier? Then check Kumo's history. If there is no call, investigate model selection and the connection path; absence alone does not prove a gateway failure.
If Kumo offers the model but your Cursor version does not support it, the next step is finding a compatible client for that API. For example, you can investigate an editor extension with configurable providers and models. Follow a dedicated recipe for each tool in Kumo's integration documentation.
Paying for GPT from Russia without a foreign card
With Kumo, you fund your own balance using a Russian bank card or SBP; crypto payment is available when offered in your dashboard. Supported models share a balance, and each request's cost depends on the selected model's rate and actual usage.
This preserves a familiar budgeting approach: allocate a monthly amount, test a few work tasks and estimate your spending pace. During intensive work, you can top up. There is no need to wait for a subscription allowance to reset, although API throughput, budget and context constraints remain. Terms are covered in the billing section and limits documentation.
How to avoid funding an experiment blindly
- Connect one compatible GPT model and start with a small top-up.
- Try three real tasks: explaining a function, locating a bug's cause and suggesting tests.
- Compare answers and token usage in Kumo's history. A long conversation and a short new chat can cost different amounts because of context volume.
- After checking, move everyday tasks to the selected connection method. Save the model identifier and address settings, keeping the key separately.
A workflow handles news more calmly when you understand where answers come from and who bills you. With Kumo, one test request and a history entry establish that before you trust the connection with a major refactoring task.
Keep coding in your familiar environment
Create a Kumo key, connect a compatible GPT model and test it on your project. That gives you a verified option today; follow Cursor's current changes as they are published.
If this is your first connection, start with the step-by-step guide to Cursor in Russia without a VPN. If you are choosing models for larger tasks, explore Opus and Fable pricing and savings calculations.
Frequently asked questions
- Has GPT already disappeared from Cursor?
- This article describes the situation on September 10, 2026. An announcement of a future end to a direct agreement does not mean every access method has already been disabled.
- Did Altman personally say he would remove GPT from Cursor?
- The verified primary source is an OpenAI corporate statement. Attributing its wording to Sam Altman as a verbatim personal quote would be inaccurate.
- Will Kumo work in Cursor after November 12?
- That depends on Cursor's future support for custom keys and compatible models. Another product's decisions cannot be guaranteed in advance.
- Do I need a ChatGPT subscription for a Kumo key?
- No. Your Kumo account and balance are separate; a ChatGPT subscription does not pay for calls through Kumo.
- Why is GPT listed in the catalogue but Cursor returns an error?
- Check the identifier, base URL and supported mode. A model's availability from an API provider does not imply compatibility with every client and every editor feature.