OpenAI Select Partner: what changes for your AI build

Creuto is now an OpenAI Select Partner, named within the OpenAI Partner Network on 21 September 2026. If you are deciding who builds your AI product, the useful question is narrower than the headline: which OpenAI models we build on today, what the status changes in delivery, and what it leaves exactly where it was, which is your data, your keys and your architecture. This post answers all three, with every model claim checked against OpenAI's developer documentation.

• Announced: 21 September 2026, from Bhubaneswar, India.
• Programme: the OpenAI Partner Network, which our announcement describes as a global programme for partners to build, sell and deliver AI solutions with OpenAI.
• Markets: India, the UAE, the United States, Australia and New Zealand.
• What we build on: GPT-6 Astra, the GPT-5.6 family (Sol, Terra and Luna), the Agents API and GPT-Live 1.

What an OpenAI Select Partner is, as far as we can source it

Our announcement describes the OpenAI Partner Network as a programme that brings together partners with industry expertise, delivery capability and customer relationships, and equips them with resources, enablement and support to help enterprises adopt OpenAI's frontier models and products. Being named a Select Partner places Creuto inside that network.

That is where our description stops. OpenAI's own partner pages were not readable to us when we wrote this, so we will not describe tiers, entry requirements or benefits we cannot point to. If any firm, including us, tells you what the badge guarantees, ask for the OpenAI page that says so.

The OpenAI models we build on, as of September 2026

OpenAI's model catalogue now leads with four text models, and the prices below are its published Standard rates per million tokens. The right-hand column is our judgement, not OpenAI's.

Model	OpenAI's description	Input / output per 1M tokens	Where we reach for it
GPT-6 Astra	Most capable model, built for the hardest end-to-end work	$10 / $50	Long agentic tasks, computer use, hard coding and document work
GPT-5.6 Sol	Flagship model for complex professional work	$4 / $20	Complex product features where Astra is more than the task needs
GPT-5.6 Terra	Balances intelligence and cost	$2 / $12	Most production features that run at volume
GPT-5.6 Luna	Optimised for cost-sensitive, high-volume workloads	$0.20 / $1.20	Classification, extraction and routing

GPT-6 Astra was released on 3 September 2026, and it is not a drop-in swap. It does not support the none reasoning effort, does not accept custom temperature or top_p, and tool calling requires the Responses API rather than Chat Completions. Two price notes matter for budgeting: GPT-5.6 Sol's current rates are promotional and available at least through 21 November 2026, and prompts over 272K input tokens are billed at 2x input and 1.5x output for the whole request.

Agents API and GPT-Live 1

The Agents API entered public beta on 10 September 2026. It gives your application OpenAI's managed Codex harness, with OpenAI handling sessions, orchestration, context compaction and recovery, and the agent running in an OpenAI-hosted sandbox or one you connect. Whether to rent that harness or keep your own is a real trade-off, which we worked through in buying the harness versus keeping yours.

GPT-Live 1 became generally available the same day. It is a full-duplex voice model that listens while it speaks and delegates reasoning and tool use to a backend model. Sessions cost $0.05 per minute, billed per second, with backend model and tool usage charged separately.

Our announcement also names ChatGPT Work, alongside GPT-5.6, as part of how we help organisations turn ambitious goals into finished work. It is a ChatGPT product rather than an API model, and this post sticks to what OpenAI's developer documentation covers, so we are not describing its features here.

What the OpenAI Select Partner status changes in delivery

The announcement commits us to continue working with OpenAI to help organisations build, deploy and scale AI solutions, to invest in team training and enablement on OpenAI models and products, and to develop reusable accelerators. The target is the step most AI projects stall on: getting out of the pilot and into production systems that are secure, scalable and maintainable.

"Being named an OpenAI Select Partner is a strong validation of how we build. Our customers are no longer asking whether AI works. They are asking how to put it into a product, integrate it with their operations, and keep it reliable at scale. Working closely with OpenAI lets us shorten the path from idea to production and turn AI investment into outcomes our customers can measure."
Nihar Rout, CEO and Co-Founder, Creuto

In practice, production means tracking the platform as it moves. On 23 October 2026 OpenAI shuts down gpt-4, gpt-3.5-turbo, o1, o3-mini and o4-mini among others. In September alone it added expiry dates for project API keys and organisation-level controls over which keys can be created. None of that shows up in a demo, and all of it decides whether an AI feature is still working a year later. That is the work our AI engineering services are built around.

What does not change: your data, your keys, your vendor choices

A fair objection to any partner badge is that the partner now has a reason to recommend that vendor. We take the objection seriously, and our answer is structural rather than a promise.

• Your account stays yours. We build into your OpenAI organisation, projects and keys, so billing, usage data and access stay under your control when an engagement ends.
• Model choice is argued with evidence. We also build with Claude and Gemini, and sometimes a trained model beats any API; we covered when in custom machine learning versus an API. Ask us to justify a model choice with evaluations on your data.
• Prices are OpenAI's published prices. Our announcement mentions no discounts and we do not claim any.

Where your data is processed in India, the UAE and Australia

Four of our five markets have OpenAI data residency options, and they are not equivalent. OpenAI's data controls guide is explicit that support for regional storage does not imply support for regional processing.

Region	Regional storage	Regional processing
United States	Yes	Yes
India	Yes	No
Australia	Yes	No
United Arab Emirates	Yes	Yes, for a subset of models including GPT-5.6 Luna

The UAE row has the catch. Regional processing there covers GPT-5.6 Luna, GPT-5.5 and GPT-5.2, not Sol, Terra or Astra, and selecting the UAE requires additional approval. Any non-US region also requires approval for abuse monitoring controls and a Modified Retention amendment, and residency endpoints carry a 10% uplift for models released on or after 5 March 2026. New Zealand has no region in the table. If in-country processing is a hard requirement, it can decide the model before any benchmark does, which is a conversation our Dubai practice has early.

The first decision for most teams

If your product still calls a 2023-era model, the partnership is less urgent than the calendar: 23 October is 32 days away, and a model that returns an error that day cannot be negotiated back. Start with an inventory of every model ID your code sends, then decide which move from OpenAI's substitute list you can prove with your own test cases.