OpenAI's proposed cutoff for its models in Cursor is November 12. Here's what to check and what to switch to
November 12 is the date OpenAI proposed for ending its models in Cursor, following SpaceX's takeover of Cursor's parent company. If your team uses GPT-5.6 Sol, Terra, or Luna in Cursor, pick replacements now, test them on real tickets, and check which features your own OpenAI key still covers.
OpenAI announced the wind-down on August 28, in a post titled “Our decision on Cursor following its acquisition by SpaceX”. As of this morning, neither company had confirmed whether the date holds. Cursor's models page still lists three GPT-5.6 models with no end date, and its changelog hasn't mentioned the change.
We haven't tested the replacements below on our own code. The prices are from Cursor's docs, and the statements are quoted from coverage we link to, because we couldn't load OpenAI's pages directly.
What happened
SpaceX completed its acquisition of Anysphere, the company behind Cursor, on August 14. Two weeks later, OpenAI said it would use a change-of-control clause to end its contract. Per Digital Applied's summary of the post, OpenAI wrote that it “cannot be confident that SpaceX will use our technology within our terms of service.”
- The date. OpenAI said it would “hold the contract cancellation to the latest date we can,” with a proposed shutoff of November 12, 2026.
- No new models. OpenAI said it would not provide future models to Cursor in the meantime, so the GPT options in Cursor stayed frozen.
- It could come sooner. OpenAI's help article says “Cursor may choose to end access sooner.”
- Cursor's response. CEO Michael Truell said OpenAI models “serve about 5% of Cursor user traffic” and that the company was “speaking with the OpenAI team to resolve this.”
- Anthropic's response. Co-founder Tom Brown posted support for Cursor users, calling Cursor “a trusted partner of Anthropic since Sonnet 3.5.”
We covered the acquisition itself in August. See our SpaceX and Cursor piece for the deal and what it means for Cursor's direction.
Who is affected
Cursor's models page lists GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna. They're third-party models outside the included usage pool, available on Premium plans and above and billed at API list price from prepaid credit. On Teams and Enterprise, they also carry Cursor's $0.25 per million token rate.
You're exposed in three ways:
- You pick a GPT model. The obvious case. Anyone who set GPT-5.6 Sol as their agent model, or pinned Luna for cheap edits, loses it.
- Your team uses Auto in Balance or Intelligence mode. Cursor's pricing docs say that on Teams and Enterprise, “Cursor Router picks among Cursor models and third-party models” for those requests. If some of those requests went to GPT models, your Auto results may shift without anyone changing a setting.
- Your automation names a model. Cloud agents, automations, and Agent SDK scripts that hardcode a GPT model ID will start failing once it's gone. Search your configs now.
Truell's 5% figure is an average across all users. If your team standardized on GPT-5.6, your share is 100%.
Does your own OpenAI key still work?
Partly. Cursor's API keys page says “Custom API keys only work with chat models” and that “Tab completion continues using Cursor's built-in models.” You add one under Cursor Settings, Models, OpenAI.
| Cursor feature | Works with your own OpenAI key? |
|---|---|
| Chat and Agent in the editor | Yes, for OpenAI chat models |
| Tab completion | No, it uses Cursor's own models |
| Auto | No, per OpenAI's help article as quoted in coverage |
| Cloud and background agents | No, per the same article |
Two catches. Requests made with your own key bill to your OpenAI account at API rates, and a ChatGPT subscription doesn't cover them. Cursor's page also says Zero Data Retention doesn't apply to those requests; your data follows OpenAI's policy instead. Neither company has said whether OpenAI will treat key-based traffic from Cursor differently after the cutoff.
What to switch to
Cursor's pricing table makes the swap easier than it sounds, because every GPT-5.6 tier has a model at about the same price. Prices are per million tokens, input and output:
| If you used | Price | Closest by price | Their price |
|---|---|---|---|
| GPT-5.6 Sol | $4 / $20 | Claude Opus 5.5, Muse Wonder | $4 / $20 |
| GPT-5.6 Terra | $2 / $12 | Claude Sonnet 5.5, Gemini 4 Carbon, Gemini 3.1 Pro | $2 / $10, $2 / $10, $2 / $12 |
| GPT-5.6 Luna | $0.20 / $1.20 | Claude Haiku 5.5, Composer 2.5 | $0.10 / $0.50, $0.50 / $2.50 |
One catch in the cheap tier. Cursor's Haiku 5.5 page says requests above 100k input tokens bill at 5x, so $0.50 / $2.50. Agent runs that carry a large context can cost more on Haiku than Luna's listed rate.
Price is only a starting point. Opus 5.5 is Anthropic's recommended default for coding, and Claude Fable 5.2 is in Cursor too at $8 / $40 if you need more capability. Our Fable 5.2 guide covers when that price pays off.
Grok 4.7 is the cheap wildcard at $2 / $6, and it draws from Cursor's weekly pool before billing at that rate. Given who owns Cursor now, expect Grok and Composer to get the most attention in the product. That's a reason to test them, and also a reason not to depend on them alone.
If you need to keep GPT
Some teams have prompts, rules files, or review habits tuned for GPT models, and switching costs real time. You have two ways to keep GPT next to Cursor:
- Bring your own key for Chat and Agent, as above. It covers interactive work and nothing that runs unattended.
- Install the Codex IDE extension. OpenAI's Codex IDE docs list Cursor as a supported editor. Codex runs in its own panel and doesn't change the model behind Cursor's Tab, Agent, or Auto.
If most of your work is GPT-based agent runs, a third option is to move those runs to Codex CLI or another editor entirely and keep Cursor for the parts you like.
What to do today
- Find your GPT dependencies. Search rules files, cloud agent configs, automations, and Agent SDK scripts for
gpt-5.6model names. - Pick one replacement per tier from the table, and run five tickets you've already closed through it. Compare the diffs against what shipped.
- Retune prompts. Rules written around one model's habits often need edits for another. Budget an afternoon, not a sprint.
- Decide on your own key before the models disappear, and tell your security team that ZDR doesn't cover those requests.
- Treat this as a drill. Any provider can leave any editor. Keep at least two models you've tested on your own code, so the next cutoff costs a settings change.
Sources
- OpenAI: Our decision on Cursor following its acquisition by SpaceX (August 28, 2026). We couldn't load it, so quotes come via the coverage below.
- Digital Applied: OpenAI is cutting off Cursor's access to its models (updated August 30, 2026): OpenAI, help center, Truell, and Brown quotes.
- InfoWorld: Cursor customers will lose access to OpenAI coding models in November (August 31, 2026): timeline and analyst views.
- Cursor docs: Models and Models and pricing: current model list, prices, plan availability, and Auto behavior.
- Cursor help: API keys: bring-your-own-key scope, setup, billing, and data retention.
- Cursor changelog: no mention of the wind-down as of November 12.
- OpenAI: Codex IDE extension: supported editors, including Cursor.
Researched and drafted by an AI agent from the linked sources. How posts are made.