THE ESSENTIALS IN 30 SECONDS

  • OpenAI's API documentation now lists Australia, India and Japan as regions where requests are both stored and processed in-country.
  • In-country inference covers only gpt-6-luna on the standard pay-as-you-go tier, via the Responses and Chat Completions APIs, without OpenAI-hosted tools.
  • Customers need OpenAI-approved zero data retention; regional endpoints carry a 10% uplift for eligible models released since March 5, 2026.

OpenAI has started processing API requests inside Australia, India and Japan for a subset of business customers. The change, reported on 12 November, means the model computation that produces a response can stay in the country, not only the data stored at rest. Until now, OpenAI offered storage residency in Japan and India (since May 2025) but ran inference elsewhere.

What changed

OpenAI's data controls documentation now shows Australia, Japan and India with both regional storage and regional processing, each with its own endpoint (au., jp. and in.api.openai.com). Customers create a project tied to the chosen region and send requests to that endpoint. Takehiro Ogino of OpenAI Japan flagged the Japanese launch on 11 November, according to XenoSpectrum, while Indian outlets covered the India rollout on 12 November.

Nitin Bawankule, OpenAI's head of enterprise for India, told People Matters that local inference gives companies "more control over where supported workloads run". The outlet notes that financial services firms in particular face localisation requirements.

The conditions

The scope is narrow. According to the documentation, in-country processing in these three regions works only with gpt-6-luna, on the standard pay-as-you-go tier, through the Responses and Chat Completions APIs. Text and image inputs are supported with text output, though images in non-US regions need separate enhanced approval. OpenAI-hosted tools are excluded, and Dataquest reports that Batch and Flex processing are not available in India.

Organisations must have active zero data retention (ZDR) approved by OpenAI; modified abuse monitoring alone does not qualify. Non-US regions also require signing a modified retention amendment, and eligibility goes through OpenAI's sales team. Regional endpoints cost 10% more for residency-eligible models released on or after 5 March 2026.

Limits and context

System data such as account details, usage records and billing information is not covered by the regional setting, so some information still leaves the country. The rollout applies to the direct API, not to ChatGPT.

The three countries join the US, Europe and the United Arab Emirates as regions where OpenAI lists in-region processing. Canada, Singapore, South Korea and the UK remain storage-only, meaning inference may happen abroad. Dataquest links the India move to OpenAI's February "OpenAI for India" plans with Tata Group for local data centre capacity, but OpenAI has not said which facilities serve these requests.