started · updated
OpenAI enables domestic inference for GPT-6 Luna in Japan
OpenAI has expanded its data residency capabilities by allowing the GPT-6 Luna model to perform inference within Japan. While OpenAI previously allowed customers to specify Japan as a data storage location starting in May 2025, this update enables the actual processing of inputs and generation of responses to occur on domestic servers.
To utilize domestic inference, users must use the Standard pay-as-you-go service tier and connect via the regional endpoint jp.api.openai.com. The feature is currently limited to the GPT-6 Luna model, which is described as a cost-effective, lightweight model. Supported interfaces include the Responses API and the Chat Completions API for text and image inputs.
Security and data retention remain critical considerations. Users must enable Zero Data Retention (ZDR) to exclude content from standard abuse monitoring logs. For certain configurations, such as image inputs, additional approvals for enhanced ZDR or Modified Abuse Monitoring (MAM) may be required. OpenAI states that API data is not used for model training without explicit consent, though data may be held for up to 30 days by default for abuse monitoring unless specific measures are in place.
Using regional endpoints incurs a 10% price premium for models released after March 2026. Additionally, while customer content can be localized, system data such as billing and account information may still be processed outside of Japan.
Entities
GPT-6 Luna · Japan · OpenAI