Hi. I configured personal token, but Cursor still do not use it:
Cursor dashboard
OPenAI dashboard
Hi Eugen, thanks for sharing the screenshots.
The charge that you’re seeing is a minimal Cursor Token fee, which applies to all requests except Auto Mode. So it applies to regular models that you choose in the model picker, as well as BYOK requests.
The Cursor Token Fee (CTF) is not the same as the model provider’s token cost; it covers Cursor-side infrastructure and product features around the request, while the provider usage is billed through your own API account.
If you’re not seeing any usage at all in your OpenAI dashboard, though, that may mean the request is not being routed through your personal key. In that case, please double-check that the key is enabled in Cursor settings and that the selected model supports BYOK.
You can read more here: Cursor Token Rate | Cursor Docs
Thank you.
Hi @Eugen_Konkov
Good questions. “Use Auto” means selecting the Auto model in the model picker. Auto is exempt from the additional Cursor Token Rate, so you won’t see that separate $0.25 / 1M token fee on Auto requests. Auto still uses Cursor’s normal request flow and has its own fixed token pricing, but Cursor chooses the model behind the scenes for each request. Auto selects the model for each request to balance intelligence, cost, and reliability, as well as regional model availability. Auto does not use BYOK.
BYOK means “Bring Your Own Key.” If you add your own OpenAI/Anthropic/Google/etc. API key in Cursor Settings > Models, Cursor can use that key for supported provider models. In that case, the provider usage is billed by your provider account, and for non-Auto agent requests, the Cursor Token Rate can still apply. The API key is sent to Cursor’s backend with the request because Cursor does final prompt building and routing, but it is not stored on Cursor’s servers.
We have more info on each of these items here:
It still not clear. Why it is excluded what is the architecture?
“Use Auto” means selecting the Auto model in the model picker.
Does this means that you can choose to send requests to China’s AIs? May we control the list of AIs where you are allowed to send such requests?
Please point all BYOK abbreviations at your documentation to the Bring your own API key | Cursor Docs page. So users will not be forced to google what BYOK means.
Hi @Eugen_Konkov, Sorry for being unclear with the Bring your Own Key lingo.
When you select Auto, Cursor chooses an available model for each request to balance quality, cost, reliability, and regional availability, and Auto has its own fixed pricing. Auto is exempt from the separate Cursor Token Rate that applies to all other requests, including BYOK requests.
Auto does not use your BYOK API key. BYOK means Bring Your Own API Key, where you add your own provider key in Cursor Settings > Models. In that case, the provider usage is billed by your provider account, and Cursor still routes the request through our backend for final prompt building. More detail is here: Bring your own API key.
On model/provider control: Enterprise customers have model access controls that can restrict which models are available to the organization. Those controls are the right path if your organization needs fine-grained allowlisting or restrictions for compliance reasons. More detail is here: Model and Integration Management. Teams plans do not currently expose the same fine-grained org-wide model allowlist controls.
Regarding your regional concerns, our models are not hosted in China, and no data is sent to Chinese servers. Our server infrastructure is located primarily in the US, Canada, and Iceland. Here’s more info about Composer: , and you can also check out our trust website for more information: https://trust.cursor.com/