OpenAI's Deprecation Calendar: Nov + Dec + Jan Shutdowns — What to Check Before Each

The October 23 shutdown — twelve OpenAI model IDs in one day — was not the end of the calendar. It was the start of a season: evals and prompts at the end of November, the first-generation GPT-5 snapshots in December, and the GPT-4o batch in January. This is the full Nov–Jan calendar, every date verified against OpenAI's deprecations page, with the specific check to run before each one lands.

The calendar, on one screen

One rule runs through every row: fine-tunes die with their base model. If you fine-tuned any ID on this calendar, your shutdown date is its shutdown date.

Run this audit once, before any of the dates

The obvious approach — grepping every repo for thirty deprecated IDs — is also the wasteful one. The reader-contributed workflow is better:

Step 1 — Dashboard first, filtered by model. The OpenAI usage dashboard, filtered by model, lists the IDs your organisation actually called. Real exposure is usually three or four IDs, not thirty.

Step 2 — Grep only for the live dying IDs on your list. Intersect the dashboard list with the calendar above, then search app code, infra configs, env examples, CI variables and scheduled jobs for just those IDs.

Step 3 — Make it permanent. OpenAI's deprecations page notes that GET /v1/models exposes a shutdown_date field per model. A monthly poll turns the next deprecation announcement into a calendar entry instead of a surprise.

November: the platform features die (Oct 31 + Nov 30)

Oct 31 — evals go read-only. After this date existing evals can be viewed but not run or modified. Eval configurations and results worth keeping should be exported in October, not November.

Nov 30 — three shutdowns on one day. The Evals dashboard and API shut down entirely (OpenAI's deprecations page points to a Promptfoo migration guide). The v1/prompts API and reusable prompt objects shut down — code referencing stored prompt IDs starts failing, and OpenAI's migration advice is to move prompt content into application code. Agent Builder also shuts down; ChatKit remains available, with the Agents SDK or ChatGPT Workspace Agents as the documented paths.

Check before Nov 30: search for pmpt_ prompt IDs and any /v1/evals or /v1/prompts calls, and inventory anything built in Agent Builder.

December: image models, then the first GPT-5s (Dec 1 + Dec 11)

Dec 1 — gpt-image-1-mini, gpt-image-1.5 (including the 2025-12-16 snapshot) and chatgpt-image-latest are removed; the replacement for all three is gpt-image-2.5-sunburst or gpt-image-2.5-flare. gpt-image-1 itself died in the Oct 23 batch. Image-model swaps change output style — if users see the generated images, review the new model's output before committing.

Dec 11 — the first-generation GPT-5 family and o3, announced June 11, 2026:

The bare aliases are listed alongside the snapshots — calling plain gpt-5 does not avoid the shutdown, and pinned August 2025 snapshots are explicitly on the list.

Check before Dec 11: swap, then replay real production prompts against the replacement and diff outputs — a generation jump is not a drop-in even when parameter shapes line up.

January: voice first, then the big one (Jan 6, 14, 20)

Jan 6 — text-to-speech: tts-1, tts-1-hd and gpt-4o-mini-tts (both snapshots) are removed, replaced by gpt-realtime-2.1-mini. The same day, creating new fine-tuning jobs closes for all customers, including active existing ones — inference on existing fine-tunes continues until the base model is deprecated.

Jan 14 — the GPT-4o batch, announced October 14, 2026:

GPT-4o was the default model for a long stretch, and defaults get embedded — in services, cron jobs and config files nobody has opened since. Fine-tuned versions of all four follow the same date.

Jan 20 — the legacy realtime/audio batch: gpt-realtime (and its 2025-08-28 snapshot), gpt-4o-realtime, gpt-audio, gpt-4o-audio, gpt-realtime-mini, gpt-audio-mini, gpt-4o-mini-realtime, gpt-4o-mini-audio and the March 2025 gpt-4o-mini-transcribe snapshot — replaced by the gpt-realtime-2.1 and gpt-audio-1.5 families.

Check before January: treat Jan 6–20 as one migration window. TTS first (small and contained), the GPT-4o swap as the main event, realtime last — fewest call sites, most behaviour to re-test.

Frequently asked questions

When does the Evals API shut down? Read-only from October 31, 2026; the dashboard and API shut down November 30, 2026 — the same day as v1/prompts and Agent Builder.

What happens to GPT-4o on January 14? Its two remaining snapshots, plus gpt-4o-mini, gpt-4.1 and gpt-4.1 mini, are removed; OpenAI's replacement for all of them is gpt-6-luna, and fine-tunes follow their base models.

Does calling the bare gpt-5 alias avoid the Dec 11 shutdown? No — the alias is listed in the deprecation table alongside the dated snapshots.

How do I audit what I actually use? Filter the OpenAI usage dashboard by model first, then grep only for the dying IDs on your own list.

Can I monitor shutdowns automatically? Yes — GET /v1/models exposes a shutdown_date field per model; poll it monthly.