Playbooks · Coding tasks
Azure retires the o-series on November 19
Written November 14, 2026
o1, o1-pro, o3, o3-pro, o3-deep-research, o3-mini and o4-mini retire in Foundry on November 19, plus Llama-4-Maverick. Microsoft's named replacements.
What retires
November 19, 2026 is the largest single retirement date on Microsoft's Foundry model retirement schedule this month. Seven Azure OpenAI models from the o-series retire that day, all currently in the Deprecated stage, and one Meta model in the same schedule retires with them. Microsoft names a replacement for every o-series row, but the replacement is not the same across the family: the full-size models point at one target and the two minis point at another. A migration script that swaps every o-series deployment to a single model will land two of them on the wrong target.
The November 19 table
- o1, version 2024-12-17 — retires November 19, 2026 — replacement: gpt-5.6-sol.
- o1-pro, version 2025-03-19 — retires November 19, 2026 — replacement: gpt-5.6-sol.
- o3, version 2025-04-16 — retires November 19, 2026 — replacement: gpt-5.6-sol.
- o3-pro, version 2025-06-10 — retires November 19, 2026 — replacement: gpt-5.6-sol.
- o3-deep-research, version 2025-06-26 — retires November 19, 2026 — replacement: gpt-5.6-sol.
- o3-mini, version 2025-01-31 — retires November 19, 2026 — replacement: gpt-5.6-terra.
- o4-mini, version 2025-04-16 — retires November 19, 2026 — replacement: gpt-5.6-terra.
- Llama-4-Maverick-17B-128E-Instruct-FP8, in the schedule's Meta section — retires November 19, 2026 — replacement: none named.
The split that matters
Five models — o1, o1-pro, o3, o3-pro, and o3-deep-research — map to gpt-5.6-sol. The two minis, o3-mini and o4-mini, map to gpt-5.6-terra. That split follows the shape of the old family: the minis were the cheaper, faster tier, and Microsoft's replacements keep them on a separate target from the full-size models. Price and latency comparisons for a mini workload should be run against terra, not sol, whatever the full-size migrations show. The Llama-4-Maverick row is the odd one out on the date: it sits in the Meta section of the schedule rather than the Azure OpenAI section, and its Replacement column is a dash, so Maverick users face the same pick-your-own-target problem as any unnamed retirement.
These are Foundry dates, not OpenAI API dates
The same model names retire on a different calendar in OpenAI's own API. The December 11 shutdown page covers the API side, where GPT-5 snapshots and o3 shut down on December 11. A team using both platforms has two deadlines, not one, and the Foundry one comes first. Do not let an API migration plan dated for December stand in for the Foundry deployments that stop on November 19.
What Retired means, and how migration runs
Under the Foundry lifecycle policy, a Retired model is removed from service: all inference requests return 410 Gone, and existing deployments no longer work. The same policy notes that provisioned deployments are not auto-upgraded, and names two migration shapes: in-place migration, where Azure handles traffic migration over a 20–30 minute window with no downtime, or side-by-side migration, where you create a new deployment, test it, switch traffic, and delete the old one. Side-by-side is the safer shape for the o-series move, because it leaves the old deployment answering while the replacement is evaluated.
What to do
List every Foundry deployment on the seven o-series versions and on Llama-4-Maverick, then group them by replacement target — sol, terra, or unnamed — before touching any of them. Evaluate each target against the workload it will inherit, with your own prompts and data. Complete the moves before November 19; after that date the old deployments return 410 Gone and there is no source left to migrate from, only a replacement to stand up from scratch.
Related playbooks
- Foundry retires codex-mini on November 15
- Anthropic's no-distillation rule lands November 12
- Copilot models after the November 16 retirement
- November shutdowns: Runloop, Conductor, Moondream
Sources
- Model retirement schedule — Microsoft Learn, Azure Foundry
- Foundry Models lifecycle and support policy — Microsoft Learn, Azure Foundry
Back to: Coding playbooks hub · Also: GPT-5 snapshots and o3 shut down Dec 11 · Foundry retires codex-mini on November 15