Lifecycle intelligence

Azure AI: the October 2026 retirement wave.

Microsoft Foundry retires six GA model versions between October 1 and 21, 2026, plus a preview. The versions actually retiring are not the ones most deployment names suggest — and the live registry currently disagrees with the official schedule on several Azure records.

Explore the live registry
Azure versions retiring
6

GA versions, Oct 1 – Oct 21, 2026

Catalog records out of step
4

Azure version groups vs official schedule

Days to first cutoff
31

Oct 1, 2026 — observed Aug 31, 2026

What Microsoft’s schedule says retires in October

Official retirement schedule for Foundry Models sold by Azure, observed August 31, 2026. The replacement column is Microsoft’s recommended replacement.

Model versionStageRetirementReplacementNote
gpt-4o 2024-05-13DeprecatedOct 1, 2026gpt-5.1Auto-upgrade on Standard SKU
o3-mini 2025-01-31DeprecatedOct 1, 2026o4-mini
o4-mini 2025-04-16DeprecatedOct 16, 2026
o1 2024-12-17DeprecatedOct 21, 2026gpt-5.6-sol
o1-pro 2025-03-19GAOct 21, 2026gpt-5.6-sol
o3 2025-04-16GAOct 21, 2026gpt-5.6-sol
gpt-realtime-2 2026-05-06PreviewAug 31, 2026Preview: short-notice retirement class

Source: learn.microsoft.com · Model retirement schedule. For gpt-4o 2024-05-13, Microsoft states the service “automatically upgrades it to gpt-5.1 on the Standard SKU in each region where that version is currently available.” learn.microsoft.com · Models lifecycle and support policy

The registry records that do not match the schedule

ALLM catalog snapshot cat_2026_08_27, observed August 26, 2026 — compared against the official schedule observed August 31, 2026.

Azure deployment recordALLM catalogOfficial schedule
gpt-4o-2024-08-06 (standard, eu/, global/, global-standard/, us/)Deprecated · EOL 2026-10-01Deprecated · EOL 2027-04-14
gpt-4.1 / gpt-4.1-mini / gpt-4.1-nano 2025-04-14EOL 2026-10-14Legacy · EOL 2027-04-14
gpt-4o-2024-05-13Available · no EOLDeprecated · EOL 2026-10-01
o1 / o1-pro / o3 / o3-pro / o4-miniAvailable · no EOLEOL 2026-10-01 → 2026-10-21

Read the catalog dates with care

For gpt-4o 2024-08-06 and the gpt-4.1 family, the registry EOL dates (2026-10-01 and 2026-10-14) are about six months earlier than Microsoft’s schedule (2027-04-14), and the registry does not yet carry EOL dates for gpt-4o 2024-05-13 or the o1/o3/o4-mini versions that retire in October. The registry evidence URL for these records (…/concepts/model-lifecycle) currently returns 404; the schedule now lives at the two pages linked in the sources below. These records are flagged for upstream correction; until then, plan with the official schedule.

Same version, different contract: Azure vs OpenAI direct

The same canonical model version can retire on different dates at each provider. Retirement dates: Azure from Microsoft’s schedule, OpenAI direct from the API deprecations page.

Model versionAzure AI (Foundry)OpenAI direct
gpt-4o 2024-05-13Oct 1, 2026 · auto-upgrade to gpt-5.1Oct 23, 2026 shutdown
gpt-4o 2024-08-06Apr 14, 2027No announced shutdown
gpt-4.1 2025-04-14Apr 14, 2027No announced shutdown
gpt-4.1-mini 2025-04-14Apr 14, 2027No announced shutdown
gpt-4.1-nano 2025-04-14Apr 14, 2027Oct 23, 2026 shutdown → gpt-5.6-luna

Sources: learn.microsoft.com · Model retirement schedule · developers.openai.com · API deprecations

What changes for an integration

  • gpt-4o 2024-05-13 on Azure Standard switches automatically on Oct 1. No manual migration, but the runtime becomes gpt-5.1: validate structured output, tool calling, latency and price deltas before the cutoff. Switching to OpenAI direct does not extend the runway — that version shuts down there on Oct 23, 2026.
  • Do not start an emergency migration off Azure gpt-4o 2024-08-06 or gpt-4.1. Microsoft’s schedule retires them on Apr 14, 2027. Any October 2026 date for those versions — including the current ALLM catalog records — does not match the official schedule.
  • o1 / o3 / o4-mini on Azure have a 6–7 week window. Cutoffs run Oct 16–21, with replacements pointing at gpt-5.6-sol and o4-mini. Test reasoning.mode behavior and schema conformance now, not in October.
  • Keep business-critical traffic off preview versions. gpt-realtime-2 retires per schedule on Aug 31, 2026, and preview models sit in a short-notice retirement class by design.

Other lifecycle moves observed this week

Officially announced or officially listed changes observed between August 24 and August 31, 2026.

OpenAI: transcription models deprecated

On August 26, 2026, OpenAI notified developers that whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe and gpt-4o-transcribe-diarize are deprecated, with removal on February 26, 2027. Recommended replacements: gpt-live-transcribe or gpt-transcribe.

Anthropic: the next direct-API retirements are “not sooner than” dates

claude-sonnet-4-5-20250929 is listed “not sooner than September 29, 2026” and claude-haiku-4-5-20251001 “not sooner than October 15, 2026”. claude-opus-4-1-20250805 retired August 5, 2026 with claude-opus-4-8 as the recommended replacement.

AWS Bedrock: Claude contracts diverge from the direct API

anthropic.claude-opus-4-1-20250805-v1:0 is still available on Bedrock with EOL January 8, 2027 (public extended access from October 8, 2026) even though the direct Claude API retired it in August. anthropic.claude-sonnet-4-20250514-v1:0 has EOL October 14, 2026. AI21 jamba-1-5-large-v1:0 has EOL November 26, 2026.

Google: no confirmed shutdown for Gemini 2.5 flash/pro yet

The official deprecation page lists “no shutdown date announced” for gemini-2.5-flash, gemini-2.5-flash-lite and gemini-2.5-pro, while the ALLM registry records 2026-10-16 for those deployments citing the same page. Treat the registry date as unverified until Google publishes one. gemini-2.5-flash-image shuts down October 2, 2026.

Groq: two Llama endpoints shut down August 16, 2026

llama-3.1-8b-instant and llama-3.3-70b-versatile are shut down, with openai/gpt-oss-20b and gpt-oss-120b / qwen/qwen3.6-27b as replacements. The deprecation applies to free and developer-tier usage; committed-spend enterprise contracts are unaffected.

Sources

All official pages observed August 31, 2026 (Europe/Paris). ALLM catalog snapshot cat_2026_08_27 observed August 26, 2026.

Related live views: models with a published end-of-life date and the OpenAI vs Azure AI deployment comparison render the current catalog records.

Deployment intelligence, explained

Why the October dates need a second look

Microsoft Foundry publishes a per-version retirement schedule. Two versions of the same model family can retire more than six months apart, and the version that retires first is not always the one a deployment name suggests.

Check the exact model version in your deployment before planning a migration. The ALLM catalog records for some Azure versions were observed out of step with the official schedule; the official schedule linked here is the planning source of record.