OpenAI's October 23 Model Shutdown: Replace These API IDs

Quick Answer

OpenAI will make a large group of older API model IDs inaccessible on October 23, 2026. The list includes o4-mini, o1, o1-pro, o3-mini, GPT-4 and GPT-4 Turbo aliases, one dated GPT-4o snapshot, gpt-4.1-nano, gpt-3.5-turbo aliases, gpt-image-1, and related fine-tuned models.

The documented replacements are not one universal GPT-5.6 alias. o4-mini and gpt-3.5-turbo move to gpt-5.6-terra; the o1, o3-mini, GPT-4, GPT-4 Turbo, and dated GPT-4o group move to gpt-5.6-sol; gpt-4.1-nano moves to gpt-5.6-luna; gpt-image-1 moves to gpt-image-2. Test the named replacement rather than replacing every string with gpt-5.6.

This is separate from the September 24 Sora and Videos API shutdown. Do not combine the two migration runbooks.

The October 23 replacement map

The current OpenAI API deprecations page publishes these mappings:

IDs that become inaccessible	Recommended replacement
o4-mini-2025-04-16, o4-mini	gpt-5.6-terra
ft-o4-mini-2025-04-16	gpt-5.6-terra
o1-2024-12-17, o1	gpt-5.6-sol
o1-pro-2025-03-19, o1-pro	gpt-5.6-sol with reasoning.mode: pro
o3-mini-2025-01-31, o3-mini	gpt-5.6-sol
gpt-4-0613, gpt-4, gpt-4-0613-completions, gpt-4-completions	gpt-5.6-sol
gpt-4-1106-preview	gpt-5.6-sol
gpt-4-turbo, gpt-4-turbo-2024-04-09, gpt-4-turbo-completions	gpt-5.6-sol
gpt-4o-2024-05-13	gpt-5.6-sol
gpt-4.1-nano, gpt-4.1-nano-2025-04-14	gpt-5.6-luna
gpt-3.5-turbo-0125, gpt-3.5-turbo, gpt-3.5-turbo-completions	gpt-5.6-terra
gpt-image-1	gpt-image-2

OpenAI also lists fine-tuned gpt-3.5-turbo, GPT-4, gpt-4.1-nano, babbage-002, and davinci-002 variants in the same shutdown wave. Their recommended landing is the corresponding Terra, Sol, or Luna base model, not a promise that the old fine-tune behavior transfers.

Alias changes are still breaking changes

Several unsuffixed aliases disappear with their dated snapshot. A code search that only looks for date strings will miss o4-mini, o1, o3-mini, gpt-4, gpt-4-turbo, and gpt-3.5-turbo.

Search more than application source. Check environment files, deployment variables, prompt-management tools, background jobs, eval fixtures, fallback arrays, observability filters, saved playground configurations, and customer-specific overrides. A forgotten fallback can remain silent until the primary model fails.

Do not “simplify” the table into one generic gpt-5.6 replacement. OpenAI gives Terra, Sol, and Luna different migration roles, and o1-pro additionally requires pro reasoning mode. Use the exact documented target as the starting point for qualification.

A replacement ID is not a completed migration

Changing the string proves only that the request reaches a surviving model. Before production cutover, replay representative work and compare:

• output quality and format stability;
• tool-call schemas and retry behavior;
• latency and concurrency under the account’s live limits;
• input, cached input, reasoning, and output token volume;
• maximum-output handling and incomplete responses;
• cost using the current official price page and the replay’s actual token mix.

Keep the old and new results side by side. A model with a lower input rate can still cost more if it emits more billed output or changes cache behavior. This article intentionally does not freeze a large price table that can move before October 23; quote from the live OpenAI API pricing page when the migration test runs.

The Completions aliases in the shutdown list need extra attention. A modern replacement model and an old endpoint are separate compatibility questions. Treat each *-completions entry as an endpoint migration, not just a model rename.

Fine-tunes need a behavior plan

The deprecations page says inference on listed fine-tuned variants ends with this wave. A recommended replacement base model does not carry an adapter, training set, or output contract forward automatically.

For each fine-tune, keep the evaluation set and compare the recommended base model with an explicit prompt or retrieval layer. If the behavior cannot be reproduced, decide whether to redesign the feature, use another supported customization path, or retire it. Buying more API credit does not preserve access to a retired model ID.

Keep nearby shutdowns separate

OpenAI’s page places other product deadlines nearby:

• 31 October 2026: existing Evals become read-only.
• 30 November 2026: the Evals dashboard and API shut down.
• 30 November 2026: reusable prompt objects and the v1/prompts API shut down.
• 30 November 2026: Agent Builder shuts down; ChatKit remains available.

Those dates can affect how a team validates the model migration, but they do not move the October 23 model deadline. Export or relocate eval and prompt assets while they are still editable.

The Sora and Videos API shutdown remains its own September 24 event with a blank replacement column. It is linked here only to prevent the two inventories from being merged.

Migration checklist

1. Search every runtime and configuration store for the IDs in the table.
2. Map each hit to OpenAI’s exact recommended replacement.
3. Replay real traffic and compare output, tokens, latency, limits, and current price.
4. Test endpoint compatibility for every Completions alias.
5. Rebuild fine-tuned behavior explicitly; do not assume it transfers.
6. Ship the new ID and a tested fallback before October 23.
7. Move Evals, prompt objects, and Agent Builder dependencies before their later deadlines.

Sources

• OpenAI API deprecations — October 23 model and fine-tune mappings plus Evals, reusable prompts, and Agent Builder dates; checked 28 August 2026.
• OpenAI API pricing — live rates to use during replay; checked 28 August 2026.