OpenAIGPT models for your agents, run where your data already lives.
OpenAI's GPT models can run three ways: through the OpenAI API, through Azure OpenAI in Microsoft Foundry, or on Amazon Bedrock. For most companies already on Microsoft, Azure keeps the models under your own subscription, contract and sign-in. As an OpenAI Select Partner, QServices builds on both and helps you choose.
Five questions. The models are the same family; the contract, data path and sign-in are not.
Do you already run Microsoft 365, Dynamics or Azure?
Must access go through Entra ID and private networking?
Would you rather have one contract and one bill with Microsoft?
Do you need data processed in a set region, such as the EU?
Do you need a feature that launches on OpenAI first, like the Agents API?
Answer the questions to see what fits.
Your prompts stay out of OpenAI's hands, access runs through Entra ID, and it bills on your Azure subscription. This is where most of our clients start.
Book a two-week assessmentNothing rules either route out. We run the same documents through both and compare accuracy, cost and the features you need.
You need features as soon as OpenAI ships them. Plan for OpenAI's residency approvals and a separate contract.
| Azure OpenAI (Microsoft Foundry) | OpenAI API | Amazon Bedrock | |
|---|---|---|---|
| Your data | Not available to OpenAI; Azure models don't call OpenAI's services | Not used for training by default; logs kept up to 30 days | Governed by AWS; not used for training |
| Where it's processed | Global, a Data Zone (US, EU or APAC) or your region | US and EU in-region; other regions store data only, with approval | AWS regions |
| Sign-in and network | Entra ID and private networking | API keys, projects and admin controls | AWS IAM |
| Billing | Your Azure subscription | An OpenAI account | Your AWS bill |
| New features | Usually after OpenAI; Global deployments first | First | Selected models, such as GPT-6 Astra since 8 September 2026 |
From Microsoft Learn, OpenAI's developer documentation and AWS, October 2026. Azure per-token prices are on the Azure pricing page.
| Model | Price per million tokens (in / out) | Best at | Where we use it |
|---|---|---|---|
| GPT-6 Astra | $10 / $50, more for prompts over 272,000 tokens | Complex reasoning and long documents | Hard exceptions and multi-step agents |
| GPT-6.1 Sol | $2 / $10 | Quality and cost in balance | Matching, drafting replies and journals |
| GPT-6 Luna | $0.10 / $0.50 | High-volume, cost-sensitive work | Reading and sorting documents and emails |
OpenAI API list prices, October 2026. Azure and AWS bill their own prices.
OpenAI's own guidance is to have people review outputs before use, especially when the stakes are high. Our agents are built that way from the first line.
Structured Outputs return exactly the fields in our schema, or the run stops.
Function calls look up the PO, receipt and supplier. Nothing is written.
The run stops before any change and puts the draft in a queue with the evidence.
Your AP lead or controller posts it, or sends it back.
Accuracy and cost per document, reported every month under one SLA.
Work that can wait up to 24 hours, like an overnight invoice run, costs 50% less.
Instructions repeated at the start of every call cost far less once cached.
On GPT-6 Astra, prompts over 272,000 tokens are billed at twice the input rate.
OpenAI's regional endpoints cost 10% more for models released since March 2026.
If you already run Microsoft 365, Dynamics or Azure, Azure OpenAI in Microsoft Foundry keeps the models under your Azure subscription, Microsoft's contract and Entra ID. Choose OpenAI directly when you need features that launch there first.
Not by default. OpenAI doesn't train on API data, and Microsoft states that prompts and replies on Azure aren't available to OpenAI or used to improve its models.
On the OpenAI API, abuse-monitoring logs and stored responses are kept up to 30 days unless zero data retention is approved. On Azure, stored items such as files and batch jobs stay in your Azure geography until you delete them.
Yes. Azure Data Zone EU deployments process data inside the EU Data Boundary. OpenAI also offers in-region processing in Europe, with its approval and a 10% price uplift on newer models.
Usually two: GPT-6 Luna to read and sort documents at volume, and GPT-6.1 Sol for matching and drafting. GPT-6 Astra is for the hardest cases. We test on your own files first.
It can call your ERP's API, but we stop it at a draft. A named person approves anything that moves money, and every step is logged.
Yes. QServices is an OpenAI Select Partner, as well as a Microsoft Solutions Partner and an Anthropic Claude Partner. See also Claude for business.
Facts on this page checked on 23 October 2026.
Founder and CEO
Chief Sales Officer
Join our live webinar on Sept 24 and see five
ways to automate meetings, approvals, and
document search with tools already in your
tenant.
