Playbooks · Coding tasks
Foundry retires codex-mini on November 15
Written November 14, 2026
codex-mini (2025-05-16) retires November 15 with no replacement named, and two gpt-chat-latest previews follow on November 17. What Foundry's schedule says.
What retires
Microsoft's Foundry model retirement schedule lists one Azure OpenAI model retiring on November 15, 2026: codex-mini, version 2025-05-16, currently in the Deprecated stage. The Replacement column for that row is a dash. Microsoft names no successor for this deployment, and the schedule offers no automatic upgrade target for it. Two more rows follow two days later, on November 17: gpt-chat-latest preview versions 2026-05-05 and 2026-06-24, both also with a dash in the Replacement column. Anyone running codex-mini in Foundry has a decision to make before tomorrow, and the schedule does not make it for them.
The three rows
- codex-mini, version 2025-05-16 — Deprecated — retires November 15, 2026 — replacement: none named.
- gpt-chat-latest, version 2026-05-05 — Preview — retires November 17, 2026 — replacement: none named.
- gpt-chat-latest, version 2026-06-24 — Preview — retires November 17, 2026 — replacement: none named.
The schedule's own introduction makes one point worth carrying into planning: you do not need to wait for a replacement to be listed before evaluating or moving to another model, and a replacement should be judged with your own application and data, comparing quality, latency, and cost. For codex-mini that is not a workaround Microsoft added for this row; it is how the schedule says every retirement should be handled. The absence of a named successor means the choice is yours, not that no move is possible.
What Retired means
The Foundry models lifecycle page defines the end state plainly. A Retired model is removed from service: all inference requests return 410 Gone, no new deployments can be created, and existing deployments no longer work. November 15 is therefore not the start of a wind-down for codex-mini. It is the day calls to the deployment begin returning errors. The Deprecated stage the model sits in now is the working window — existing customers can still create and manage deployments, and existing deployments still serve traffic.
The two stages in play
codex-mini sits in the Deprecated stage, which the lifecycle page defines by subscriptions: existing customers can continue to create and manage deployments, while new customers cannot create deployments or access the model, and whether a subscription counts as existing turns on whether it has ever deployed that model version. The gpt-chat-latest rows sit in Preview, and preview models run on different rules. The lifecycle page says a retiring preview is either force-upgraded to a replacement or retired with no replacement, and there is no option to remain on a retiring preview model — all preview deployments are either upgraded or terminated. With a dash in the Replacement column for both November 17 rows, termination is the outcome to plan for.
A conflict worth one line
Some roundups of this schedule paired Claude Haiku 4.5 with the November 15 date. The live schedule does not: its claude-haiku-4-5 rows are in GA with retirement dates in February 2027. Plan migrations on the schedule as it stands now, not on the older snapshot those roundups used.
What to do
Inventory first. Find every Foundry deployment of codex-mini 2025-05-16 and every deployment pinned to the two gpt-chat-latest preview versions, including test and staging resources, which fail just as hard on the day. Pick a target model per workload and run it against your own prompts before the date; with no named replacement, a same-day swap with no evaluation is the risky option. Finish the codex-mini move before November 15 and the gpt-chat-latest move before November 17. After those dates there is nothing to migrate from — the deployments stop answering, and the work becomes a rebuild under time pressure instead of a planned switch.
Related playbooks
- Azure retires the o-series on November 19
- Codex 0.171 turns on the TUI workspace
- OpenAI buys Pierre: Code Storage comes to Codex
- Cloudflare runs the Codex loop in a Durable Object
Sources
- Model retirement schedule — Microsoft Learn, Azure Foundry
- Foundry Models lifecycle and support policy — Microsoft Learn, Azure Foundry
Back to: Coding playbooks hub · Also: Re-pin Copilot models before retirements · Azure retires the o-series on November 19