OpenAI Brings GPT-6 Luna API Processing to India — With a 10% Regional Premium

OpenAI now processes GPT-6 Luna API requests in India, for customers with active Zero Data Retention, through the direct API only, via the India regional endpoint in.api.openai.com, at a 10% premium over the model's standard rates. The rollout, on 12 November 2026, extends OpenAI's existing India data residency — which covered storage — to processing, so that for supported workloads the API request and the model computation that generates the response take place in India. The scope is narrow, the conditions are specific, and the price is stated.

The distinction at the heart of the announcement is between storage and processing. Storage is where data rests. Processing is where the work happens — where the API request is received and the model computation that generates the response takes place. OpenAI's own warning is worth keeping in mind throughout: "Support for regional storage does not imply support for regional processing." The two are different, they are offered separately, and a customer who has one does not automatically have the other. This article is about the second of the two.

One model, two APIs, direct API only

At launch, only one model is covered: GPT-6 Luna, identified in OpenAI's documentation as gpt-6-luna. OpenAI states: "In Australia, India, and Japan, organizations with active Zero Data Retention can use gpt-6-luna for regional processing." That sentence names the model, the three regions — Australia, India and Japan — and the condition, active Zero Data Retention. No other model is named in the verified material for this article, and no other model is added here.

The API scope is equally narrow. Only the direct API is covered, and only two of its interfaces: the Responses API and the Chat Completions API. OpenAI states: "Use a project in the selected region. The Responses and Chat Completions APIs support text and image inputs and text output. Hosted tools are unavailable for regional processing in these three regions." Text and image inputs are supported, and text output is supported. Hosted tools are not available for regional processing in these regions. ChatGPT is not covered. This is an API-only change, not a change to the consumer product.

ItemWhat applies in India
ModelGPT-6 Luna (gpt-6-luna) only at launch
APIsResponses and Chat Completions only
Inputs and outputsText and image in; text out
Hosted toolsUnavailable for regional processing
ConditionActive Zero Data Retention required
Processing modeStandard pay-as-you-go only
Endpointin.api.openai.com
PriceStandard rates plus a 10% regional premium
Modern data centre in India with rows of server racks, an engineer in a hi-vis vest checking equipment
Modern data centre in India with rows of server racks, an engineer in a hi-vis vest checking equipment (AI-generated illustration, credit: INORYVA)


Zero Data Retention is required — and it is not automatic

Regional processing in Australia, India and Japan requires active Zero Data Retention, known as ZDR. OpenAI states the rule without qualification: "Regional processing in Australia, India, and Japan requires active Zero Data Retention (ZDR). Modified Abuse Monitoring (MAM) alone is insufficient." Modified Abuse Monitoring is a different arrangement, and it is not enough on its own. A customer with Modified Abuse Monitoring but without active Zero Data Retention does not qualify for regional processing in these three regions.

Zero Data Retention is also subject to OpenAI's approval. It is not a switch a customer can simply turn on and assume will be granted. This article does not state how many customers are eligible, how many have been approved, or how long approval takes, because none of those figures was disclosed. Under ZDR, the store parameter for the Responses endpoint and the Chat Completions endpoint is always treated as false. The request is not stored, and a customer cannot override that by setting the parameter the other way.

Developer's laptop on a desk showing an API code editor, a cup of chai beside it, warm office light
Developer's laptop on a desk showing an API code editor, a cup of chai beside it, warm office light (AI-generated illustration, credit: INORYVA)


How to set it up, and the 10% premium

The setup is a short sequence. Create a new API project, select India for storage and processing, and use the India regional endpoint in.api.openai.com. The region is a property of the project, not a per-request choice. One secondary source printed a garbled version of the endpoint — "in.apo.oepnai.com" — and that is a typo. It should not be used. The correct endpoint is in.api.openai.com, as stated in OpenAI's own docs. The processing mode is also fixed: the three regions support Standard pay-as-you-go only.

The price is verified twice over, and it should not be dropped. The data-controls documentation states: "Data residency endpoints are charged a 10% uplift for models released on or after March 5, 2026, that are eligible for data residency." GPT-6 Luna qualifies. The GPT-6 Luna model page states: "Regional processing adds a 10% premium where available." The premium is a regional-processing uplift on the model's rates, and it should be described as OpenAI describes it — a 10% uplift, or a 10% premium where available — not as an India-only tax.

The base rates for GPT-6 Luna on Standard processing, per one million tokens, are input $0.10, cached input $0.01, cache writes $0.125, and output $0.50. Regional processing is those rates plus 10%. By arithmetic, that gives $0.11 for input and $0.55 for output per one million tokens. The 10% rule itself is the sourced fact. The dollar results are calculated from it, and they are presented here as calculated, not as separately quoted prices. The benefit is that the request and the computation take place in India. The cost is 10% on top of the standard rates.

Mumbai skyline at dusk with modern glass towers and warm city lights
Mumbai skyline at dusk with modern glass towers and warm city lights (AI-generated illustration, credit: INORYVA)


What this is not

It is worth gathering the limits together, because the headline alone — "OpenAI now processes data in India" — would overstate what has happened. What has happened is narrower: GPT-6 Luna, through the Responses or Chat Completions API, with active Zero Data Retention, on Standard pay-as-you-go, without hosted tools, via in.api.openai.com, at a 10% premium. ChatGPT is not covered. No other model is covered at launch. The same expansion covers Australia and Japan, under the same conditions. The rollout sits inside OpenAI's "OpenAI for India" programme, which includes a Tata Group data-centre partnership, but the sources explicitly say the link to a specific Tata facility is not established, and this article does not tie the rollout to one. No data-centre location, capacity figure or investment figure is stated here, because none was disclosed.

There is also no standalone openai.com newsroom announcement URL located for this rollout, and this article does not fabricate one. The primary sources are OpenAI's developer documentation — the data-controls guide and the GPT-6 Luna model page — and those are the sources cited below. For context on OpenAI's wider platform work, see our coverage of the Google Cloud OpenAI partnership, the OpenAI Decisions API, GPT-6.1 Terra and OpenAI offline training. None of those stories changes the narrow scope of this rollout.

Key Takeaways

  • OpenAI now processes GPT-6 Luna API requests in India, extending existing India data residency from storage to processing for supported workloads.
  • Only one model is covered at launch — GPT-6 Luna (gpt-6-luna) — and only two APIs: Responses and Chat Completions, with text and image inputs and text output.
  • Active Zero Data Retention is required; Modified Abuse Monitoring alone is insufficient, and ZDR is subject to OpenAI approval.
  • Hosted tools are unavailable for regional processing in India, Australia and Japan, and Standard pay-as-you-go is the only processing mode in those regions.
  • Setup is a new API project with India selected for storage and processing, using the endpoint in.api.openai.com.
  • Regional processing adds a 10% premium — $0.11 input and $0.55 output per one million tokens by calculation from the $0.10 and $0.50 base rates.
  • ChatGPT is not covered; this is API-only, and the rollout should not be tied to a specific Tata facility.

Conclusion

OpenAI's India API processing rollout is a precise, narrow change, and it should be reported as one. For a supported workload — GPT-6 Luna, through the Responses or Chat Completions API, with active Zero Data Retention, on Standard pay-as-you-go, without hosted tools — the request and the model computation now take place in India, via in.api.openai.com, at a 10% premium. That is useful for the developers and organisations it covers, and it is not a blanket claim about OpenAI processing data in India, or about ChatGPT, or about any other model. The checklist is short: one model, two APIs, ZDR required, 10% uplift. Everything else is background.

Sources

Comments