Azure Government o3-mini Retirement: How Do You Move to gpt-5.6-terra?

Quick answer. As of October 2026, Microsoft’s Azure Government model retirement schedule lists o3-mini version 2025-01-31 as Deprecated with a retirement date of November 19, 2026 and gpt-5.6-terra as its replacement, and Microsoft’s commercial schedule lists the same date and the same replacement. Once a model is retired every inference request to it returns 410 Gone. Microsoft’s Azure Government lifecycle policy states that provisioned deployments aren’t auto-upgraded, so each provisioned o3-mini deployment is moved by hand before November 19, 2026, and its September 2026 availability table lists gpt-5.6-terra under Data Zone Provisioned managed in usgovarizona and usgovvirginia, so the move stays in the same deployment type.

A subscription owner forwarded the retirement notice, and a common first reading is that the swap is a one-line change: point the application at a gpt-5.6-terra deployment and move on. That reading is half right. The code change can be small, but three things decide whether the date passes quietly: which deployments Microsoft will move for you, whether the same code still behaves the same after the swap, and whether anyone measured the answers before the model changed. This guide works through each, in the order a cutover plan needs them, starting from Microsoft’s own Azure Government retirement schedule.

The Notice in Your Inbox and the Row Behind It

Microsoft’s Azure Government lifecycle policy sends a retirement notice to subscription owners with active deployments at least 60 days before retirement, and a warning at least 30 days before. For a November 19, 2026 date, that places the notice on or before September 20 and the warning on or before October 20.

The row behind the notice, on Microsoft’s Azure Government retirement schedule, reads: model o3-mini, version 2025-01-31, lifecycle Deprecated, retirement date 2026-11-19, replacement gpt-5.6-terra. Microsoft states what the schedule is for: “Use it to plan migrations before a model is deprecated or retired.”

The commercial cloud’s model retirement schedule gives o3-mini the same November 19, 2026 date and names the same replacement, gpt-5.6-terra, so the target is the same in both clouds.

What Happens on November 19, and What Is Already True Before It

Microsoft’s Azure Government lifecycle policy describes the Retired stage in two short sentences: “Removed from service. All inference requests return 410 Gone.” An application still calling an o3-mini deployment after the date gets an error, not a degraded answer.

The Deprecated stage, where o3-mini sits now, already limits who can deploy it. The same policy decides who counts as an existing customer at the subscription level, by whether that Azure subscription ever deployed the specific model version, and states that a new subscription under the same tenant doesn’t inherit access. A fresh subscription stood up for side-by-side testing should not be expected to deploy o3-mini; run the comparison in a subscription that already has it.

Microsoft’s general lifecycle policy also keeps an emergency route open: “If a model is found to have compliance or security issues, Microsoft reserves the right to invoke an emergency retirement with shortened notice.” A plan that finishes on the last available day leaves no room for that.

Which o3-mini Deployments Microsoft Moves for You, and Which It Does Not

Three passages in Microsoft’s Azure Government lifecycle policy set the upgrade rules. For Data Zone Standard: “For Data Zone Standard and Standard deployment types, Microsoft manages automatic upgrades when a model version is retired and the deployment type and region align.” For provisioned: “Provisioned deployments aren’t auto-upgraded. Provisioned customers must manually migrate to the replacement model.” For any type: “If the upgrade target isn’t available in the same deployment type and region, Microsoft doesn’t perform a model upgrade.”

Read against Microsoft’s Azure Government model availability tables (page dated September 1, 2026, read October 19, 2026), those rules sort the deployment types as the planning table shows. The table is this guide’s own planning view built from Microsoft’s tables, not a Microsoft table; confirm each row in your own subscription.

Deployment type	o3-mini listed	gpt-5.6-terra listed	What Microsoft’s rules say	What to do
Data Zone Standard	Yes, in both US Gov regions	Yes, in both US Gov regions	Automatic upgrade applies where the deployment type and region align	Set reasoning_effort explicitly and test the swap anyway
Data Zone Provisioned managed	Yes, in both US Gov regions	Yes, in both US Gov regions	Provisioned deployments are not upgraded automatically	Move it by hand to gpt-5.6-terra in the same deployment type before November 19, with a named owner

Microsoft’s Azure Government model availability tables list o3-mini in neither the regional Standard tab nor the regional Provisioned tab, and Microsoft’s general lifecycle policy states: “Global Standard deployments aren’t available in government clouds.”

A provisioned o3-mini deployment is a manual cutover whatever else happens. As of the September 1, 2026 availability table, gpt-5.6-terra is listed under Data Zone Provisioned managed in usgovarizona and usgovvirginia, the same deployment type and regions where o3-mini is listed, so the provisioned capacity moves without a change of deployment type. Each move is still made by hand, before November 19, 2026.

Microsoft’s Azure Government lifecycle policy also sets general rules for replacement timing: “Replacement model becomes available to test in Azure Government where the predecessor is retiring approximately 30 days before retirement.” And: “If the Azure Government availability of the replacement is less than 30 days from this retirement date, the model’s retirement date is extended in Azure Government to allow at least 30 days of overlap in the Azure Government cloud.” Those are Microsoft’s general rules; whether either changes anything for this row is a thing to check, not to assume.

The check itself is programmatic. The same policy says: “Use the Models API to programmatically check lifecycleStatus, deprecation, and per-SKU deprecationDate for any model at any time.” Run it against every subscription that holds an o3-mini deployment, and put an owner’s name beside each result.

What Changes When the Same Code Calls gpt-5.6-terra

The deployment name is the smallest change, and it is not why gpt-5.6-terra can answer differently with the same code. Microsoft’s reasoning models guide and its Azure Government model catalog name the differences a rename misses:

• Reasoning settings vary by model. The reasoning guide states: “Supported values vary by model and include none, minimal, low, medium, high, xhigh, and max. Defaults vary by model as well.”
• Tools and reasoning split by API. The reasoning guide states: “The gpt-5.6 models support the Chat Completions API and function tools, but not both at the same time unless reasoning_effort is none. Use the Responses API for tool calling.”
• The token parameter is fixed. The reasoning guide states: “Reasoning models will only work with the max_completion_tokens parameter when using the Chat Completions API.”
• Message roles have a rule. The reasoning guide states: “You should not use both a developer message and a system message in the same API request.”
• Output format can shift. The reasoning guide states: “By default the o3-mini and o1 models will not attempt to produce output that includes markdown formatting.” Anything that parses model output should be tested for format changes after the swap.
• Inputs and limits differ. The catalog lists o3-mini as text-only, with a maximum request of 200,000 input tokens and 100,000 output tokens, and gpt-5.6-terra with text and image processing and a 1,050,000-token context window (922,000 input, 128,000 output).
• Billing follows reasoning. The reasoning guide states: “Reasoning tokens are billed as output tokens, so a request that thinks longer costs more even when the visible answer is short.” It also states: “If you upgrade an existing workload to a GPT-5.6 model, expect higher token consumption on multi-turn conversations even when your code doesn’t change.”

The planning rule that follows: set reasoning_effort explicitly for every application, record the setting beside the deployment, and treat any later change to it as a change to be tested.

Testing the Swap Starts With the Before State You Capture Now

A before-and-after comparison only means something if both sides hold still. On the commercial schedule, gpt-5.6-terra version 2026-07-09 is listed as GA, the stage Microsoft’s lifecycle policy describes plainly: “Weights and APIs are fixed.”

The before side is the one that disappears. Capture it while o3-mini still answers, and run the gpt-5.6-terra side with the settings in Microsoft’s reasoning models guide:

1. For each application on o3-mini, freeze a set of real prompts with the answers your users accepted, taken from o3-mini now.
2. Run the same set against gpt-5.6-terra at each reasoning_effort level under consideration, in a subscription that can deploy both models.
3. Compare accuracy, refusals, output format, latency and token use, application by application.
4. Record the chosen setting and the comparison as the change record for the swap.
5. Route the change through the same environment separation and change control as any other production release.

A model swap is a production change like any other. i3solutions delivery includes ALM practices with Power Platform pipelines or Azure DevOps integration, environment separation strategies, and change control processes. Whether a model change inside your authorized boundary needs anything beyond your normal change control is a question for your authorizing official and security reviewer; this guide does not answer it.

If there is no evaluation set of any kind today, that gap is larger than this retirement, and LLM adoption consulting covers building one.

Cutover, Rollback and the Next Date on the Same Table

The plan, in order: inventory each o3-mini deployment by deployment type and region, test prompts and outputs on gpt-5.6-terra, move the provisioned capacity by hand, and confirm every deployment before November 19, 2026. Write a cutover date per deployment so that a failed comparison still leaves time to fix the prompts or the setting and run it again, and keep the o3-mini deployment in place until the gpt-5.6-terra deployment has carried real traffic. Microsoft’s Azure Government lifecycle policy names where warnings surface and how to route them: “Go to Service Health > Health advisories, filter by Azure OpenAI Service, and create an alert rule for email, text messages, or webhook notifications.”

The same Azure Government schedule carries the next dates: gpt-4.1, gpt-4.1-mini and gpt-4o for April 14, 2027, three text embedding models for April 15, 2027, and gpt-5.1 for May 15, 2027. Microsoft’s general lifecycle policy adds: “Government clouds typically support only one version of a given model at a time, with a 30-day overlap when a new version becomes available.” Build the test set once and keep it for the next retirement on the same table.

If the retirement has reopened the question of where to build at all, Copilot Studio vs Azure AI Foundry covers that platform choice.

When This Plan Is Overkill, and When Waiting Is Wrong

Not every o3-mini deployment needs the full sequence. Read against Microsoft’s Azure Government lifecycle policy, three cases sort out:

• Overkill: a single Data Zone Standard deployment in low-stakes internal use, where the same type and region rule does not block an upgrade. Still set reasoning_effort explicitly and run a short comparison.
• Waiting is wrong: anything provisioned, anything whose output feeds a regulated record, and anything with no before-state measurement.
• Weighing another model: a team may also weigh gpt-5.1, which the Azure Government schedule lists as GA with a retirement date of May 15, 2027 and the availability tables list under Data Zone Standard but not under Data Zone Provisioned managed. The replacement Microsoft names for o3-mini is gpt-5.6-terra.

What a First Conversation With i3solutions Produces

i3solutions delivers through its Expert Delivery Model, a four-phase methodology (discovery, architecture, build, and knowledge transfer) with explicit exit criteria and delivery assurance at every handoff. What the record states about i3solutions and AI is this: i3solutions delivers its full service portfolio into the healthcare vertical exactly as it does into defense manufacturing and finance, including identity work with Okta and Entra ID, AI work, Power Platform, and Dynamics 365 integration. i3Solutions architects are 100% U.S.-based.

None of that is a claim that i3solutions has migrated a model deployment; the sequence in this guide is offered as a method. The wider AI practice is described at custom AI consulting services.

Key Takeaways

The short version, from Microsoft’s Azure Government retirement schedule and lifecycle policy:

• o3-mini version 2025-01-31 retires in Azure Government on November 19, 2026, and the replacement named there is gpt-5.6-terra, the same model the commercial cloud names.
• After the date, requests return 410 Gone; provisioned deployments are not upgraded automatically, so each is moved by hand to gpt-5.6-terra, which the September 1, 2026 table lists under Data Zone Provisioned managed in both US Gov regions.
• reasoning_effort values and defaults vary by model, so pass a level explicitly.
• Capture the before state now, cut over early, and keep the test set for the next retirement on the same table.

Frequently Asked Questions

When does o3-mini retire in Azure Government?

According to Microsoft Learn, o3-mini version 2025-01-31 retires in Azure Government on November 19, 2026. Microsoft Learn publishes that date with no time of day, so plan to be cut over well before it.

What does Microsoft name as the replacement, and is it the same as in the commercial cloud?

According to Microsoft Learn, the Azure Government schedule names gpt-5.6-terra as the replacement, and the commercial schedule gives the same date and names the same model.

Will my o3-mini deployment be upgraded automatically?

Not if it is provisioned. According to Microsoft Learn, provisioned deployments aren’t auto-upgraded, and no upgrade happens where the target isn’t available in the same deployment type and region; Microsoft Learn’s Azure Government table, dated September 1, 2026, lists both o3-mini and gpt-5.6-terra under Data Zone Provisioned managed in usgovarizona and usgovvirginia, so each provisioned deployment is moved by hand in the same deployment type before November 19, 2026. Check each deployment with the Models API, which Microsoft Learn says reports lifecycleStatus and per-SKU deprecationDate at any time.

What happens to requests after the retirement date?

According to Microsoft Learn, a retired model is removed from service and all inference requests return 410 Gone.

Why can gpt-5.6-terra answer differently with the same code?

According to Microsoft Learn, reasoning_effort values and defaults vary by model, so code written for o3-mini should pass a reasoning_effort level explicitly. Microsoft Learn also notes that the gpt-5.6 models support the Chat Completions API and function tools but not both at the same time unless reasoning_effort is none, that o3-mini by default avoids markdown formatting, and that reasoning tokens are billed as output tokens. Test output format, answer quality and token use before cutover.

What should a team weighing gpt-5.1 instead know?

According to Microsoft Learn, the Azure Government schedule lists gpt-5.1 version 2025-11-13 as GA with a retirement date of May 15, 2027 and no replacement named yet, and the Azure Government availability tables list gpt-5.1 under Data Zone Standard but not under Data Zone Provisioned managed. Microsoft Learn also reserves an emergency retirement with shortened notice for a model found to have compliance or security issues.

Sources

As of October 2026, from Microsoft Learn:

• Model retirement schedule in Azure Government, last updated July 28, 2026
• Model retirement schedule (commercial cloud), last updated September 23, 2026
• Foundry Models sold by Azure in Azure Government, last updated September 1, 2026
• Microsoft Foundry Models lifecycle and support policy in Azure Government, last updated July 28, 2026