The 2026-10-23 legacy snapshot shutdown day: 17 model rows go offline, with the official replacement table

Today is a scheduled execution date on OpenAI's 2026 deprecation timeline: the "Legacy GPT model snapshots" batch announced on 2026-04-22 hits its deadline, and the old GPT-3.5 / GPT-4 era snapshots, the o1 era, the first GPT-4o snapshot, and gpt-image-1 — together with their fine-tuned versions — complete a 184-day notice period. This article verifies the shutdown list and the replacement mapping row by row against the official deprecations page as fetched on 2026-10-23, with two SDK-layer cross-checks; every statement traces to a source loaded today.

1. The shutdown date and its notice period

The deprecations page section is dated 2026-04-22, with the original wording: To improve reliability and make it easier for developers to choose the right models, we are deprecating a set of older OpenAI models. Access to these models will be shut down on the dates below. Announcement to shutdown is 184 days by calendar computation (2026-04-22 → 2026-10-23). The same page states its own policy: unless safety or compliance concerns require a faster timeline, generally available models get at least 6 months' notice and specialized variants of GA models (chat / codex / deep research) at least 3 months — this batch's 184 days meets the GA floor.

One state detail worth recording: in the fetch made on 2026-10-23, the page still listed this batch in its active deprecation table with future-tense wording (will be shut down), not yet moved into the Past deprecations section. This site holds no API key and cannot verify by live calls whether the model IDs now return errors — this article makes no "shutdown has taken effect" claim; actual availability follows your platform calls.

2. The main table's 12 rows: snapshots, aliases, and official replacements

The main table, quoted row by row from the deprecations page (every Shutdown date is October 23, 2026):

Three reading notes: aliases die with their snapshots — the bare aliases gpt-4, o1, o1-pro, o3-mini, o4-mini, gpt-4-turbo, gpt-4.1-nano, and gpt-3.5-turbo all hang off snapshot rows, so migrating only the dated strings is not enough; the only specially annotated replacement is the o1-pro row, where the official substitute carries a reasoning.mode: pro suffix; and ft-o4-mini-2025-04-16 sits in the main table, not in the fine-tuned section below.

3. The 5 fine-tuned rows

After the main table, the page reads We are also removing fine-tuned versions as below, listing fine-tuned models against their recommended replacement base:

In total: 12 main rows + 5 fine-tuned rows = 17 rows (the count of "October 23, 2026" occurrences on the page is also 17, matching the row count).

4. Two SDK-layer cross-checks

• openai-python v3.37.0: the ChatModel annotation (src/openai/types/shared/chat_model.py) carries a row reading Announced shutdown date: 2026-10-23 with 16 model strings: gpt-4.1-nano, gpt-4.1-nano-2025-04-14, o4-mini, o4-mini-2025-04-16, o3-mini, o3-mini-2025-01-31, o1, o1-2024-12-17, gpt-4o-2024-05-13, gpt-4-turbo, gpt-4-turbo-2024-04-09, gpt-4-1106-preview, gpt-4, gpt-4-0613, gpt-3.5-turbo, gpt-3.5-turbo-0125 — matching the main table's text models one to one (gpt-image-1 is correctly absent, not being a ChatModel).
• openai-node v7.37.0 (released 2026-10-17): the Chores section already carried add shutdown notices for GPT-4 model aliases (#2959) — the SDK flagged GPT-4 aliases at the type and doc layer ahead of this batch.

5. A migration checklist

1. Grep your repository for every model string: run each of the 17 rows above (including all bare aliases) through code, configuration, and prompt templates; rewrite every hit per the official replacement row.
2. Follow the official row mapping — no do-it-yourself "nearest upgrade": the mapping is per row (o3-mini goes to sol, not terra); for the broader selection framework see the links at the end.
3. The o1-pro migration carries a mode: the replacement row's reasoning.mode: pro suffix means swapping the model ID alone does not reproduce the official annotation.
4. Image requests change in two places: the gpt-image-1 replacement offers two IDs (sunburst / flare) — choose by your quality/cost trade-off, and check parameter compatibility against the official image generation guide.
5. Fine-tune bindings need account-side checks: the fine-tuned rows only name replacement bases; your account's fine-tuning tasks and bindings are authoritative in the platform console, and this site cannot query them for you.

6. What remains after this batch

This is not the end of the legacy queue; as verified on the deprecations page today: 2026-10-31 existing evals become read-only; 2026-11-30 the Evals panel and API, v1/prompts, and Agent Builder shut down the same day; 2026-12-01 gpt-image-1-mini / gpt-image-1.5 / chatgpt-image-latest → gpt-image-2.5; 2026-12-11 the GPT-5 and o3 snapshots → the corresponding GPT-5.6 tiers; 2027-01-06 the TTS quartet → gpt-realtime-2.1-mini; 2027-01-14 the GPT-4o / GPT-4o mini / GPT-4.1 / GPT-4.1 mini workhorse snapshots → gpt-6-luna; 2027-01-20 the legacy audio/realtime family; 2027-02-26 the transcription quartet; and 2027-04-01 gpt-5.1 / gpt-5.3-codex / gpt-5.4-nano → the GPT-6 tiers. For a horizontal read of the whole October wave, see OpenAI's October 2026 deprecation wave: GPT-4o and GPT-4.1 on 2027-01-14, TTS to gpt-realtime-2.1-mini.

7. Next steps

• For the full October 2026 deprecation wave (including the 2027-01-14 batch and the TTS migration), see OpenAI's October 2026 deprecation wave: GPT-4o and GPT-4.1 on 2027-01-14, TTS to gpt-realtime-2.1-mini.
• For positioning and selection of the replacement models, see The complete guide to GPT models (2026-07): GPT-5.6 Sol, Terra, Luna.
• For the week's two other developer-side stories: OpenAI models on Vertex AI in OpenAI models land on Google Cloud: the Responses API on Vertex AI and the SDK vertex() provider, and the Decisions API GA in Decisions API goes GA: a fast, low-cost judgment endpoint, explained.