What's this about?
The most repeated finding in the HN thread is that OpenAI's models were a small minority of Cursor usage — the circulating estimate is around 5%. Cursor's own models (Composer, now grok-cursor) are competitive and generously included in base plans, and Claude covers most of the rest. Cursor emailed customers saying it is talking to OpenAI and will continue offering ways to access all models.
What is "bring your own key"?
Letting a user paste their own API key so the product calls the provider on their behalf, billed to them rather than to the product. Cursor supports this for OpenAI today, and it is the obvious escape hatch. Two things make it less of a solution than it sounds: without model access for prompt engineering and evaluation, Cursor cannot tune its harness for new OpenAI models, so it may simply not offer them; and OpenAI can restrict the pattern if it chooses.
What's the community saying?
- The 5% estimate is the number everyone repeats, sourced from another commenter rather than from Cursor — treat it as a plausible community estimate, not a disclosed figure.
- A dissenting voice from someone who used OpenAI models in Cursor for six months, having switched away from Claude for speed, and is surprised the share is so low.
- The commercial read that gets agreement: OpenAI is probably still subsidising inference and sees no reason to subsidise a competitor's product experience.