OpenAI's API can now keep GPT-6 Luna requests inside Japan
Japanese developers can now have one OpenAI model process requests inside the country, but only after an approval step and only for that model.
The short version
- In-country processing in Japan works only for the model gpt-6-luna.
- It covers the Responses API and the Chat Completions API, on the Standard pay-as-you-go tier only.
- Customers must contact OpenAI and be approved for MAM and ZDR data-handling measures.
- An employee at OpenAI's Japanese arm announced it on X; ITmedia reported it.
- GPT-6 Luna is a lightweight model that OpenAI positions as strong on cost for its performance.
Why it matters
Companies in Japan with rules about where data is handled now have an official option from OpenAI, at least for one model. The conditions are strict, though: if your app uses a different model or a different pricing tier, this does not apply yet. It also covers processing only, so check separately where logs and stored data sit.
What changed
OpenAI's API now allows requests to GPT-6 Luna to be processed inside Japan. As itmedia.co.jp reported, an employee of OpenAI's Japanese arm announced it on X.
The technical term is inference: the step where a trained model reads your input and produces an answer. Until now, a Japanese customer could not ask for that step to stay in the country. Japan has now been added to the list of regions where OpenAI offers it.
The fine print
The limits matter more than the headline. Only the model named gpt-6-luna is covered. Only two interfaces qualify, the Responses API and the Chat Completions API, and only on the Standard pay-as-you-go service tier.
It is also not self-service. A customer has to contact OpenAI and be approved for two measures. MAM is a modified form of abuse monitoring. ZDR is zero data retention, meaning OpenAI does not keep the content.
How it fits OpenAI's data policy
OpenAI says data sent through its API is not used to train or improve models unless the customer explicitly agrees. By default, though, it keeps prompts, model answers and input content for a limited time so it can watch for misuse.
Approved customers can have their content left out of those abuse-monitoring logs. In-country inference builds on that arrangement, which is why it comes bundled with the MAM and ZDR requirement.
Who it is for
GPT-6 Luna is a lightweight model that OpenAI presents as good value for its cost. It suits high-volume tasks more than the hardest reasoning jobs. Teams with Japanese data-handling rules could use it as a starting point.
Teams that depend on a different OpenAI model get nothing from this announcement. The report also does not say whether other models or tiers will follow.
What to watch
- Will OpenAI extend in-country inference in Japan to other models or service tiers?
- Where do stored logs and files live, which this report does not address?
Try it
If you build on OpenAI's API in Japan, check which model your app is pinned to and test whether GPT-6 Luna gives acceptable quality for that task.
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by itmedia.co.jp · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.