Model deprecation calendar, Q4 2026: Anthropic, Google and OpenAI dates

As of 10 October 2026, dozens of model IDs have dated shutdowns before the end of the year across the three big APIs, and the dates cluster: 22 and 23 October, 17 November, 30 November, 1 December, 11 December. This is the calendar, built only from each vendor’s own deprecations page, with the dates that look like retirements but are not separated out.

I compiled it from the vendor pages listed below, fetched on 10 October. Vendors edit these pages without announcement, so treat the table as a snapshot and re-check the source column before you act on a date.

What gets retired in Q4 2026?

All dates 2026. Source column: A is Anthropic’s deprecations page, G is Google’s deprecations page, O is OpenAI’s deprecations page. Every row is a vendor page; none is third-party.

Date	Vendor	What	Migrate to
22 Oct	G	veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-preview, gemini-omni-flash-preview	gemini-omni-1.1-flash
23 Oct	G	deep-research-pro-preview-12-2025	deep-research-preview-04-2026
23 Oct	O	Legacy GPT snapshots and fine-tunes (grouped below)	gpt-5.6-terra, gpt-5.6-sol, gpt-5.6-luna
17 Nov	G	gemini-3.1-flash-tts-preview, gemini-2.5-flash-preview-tts, gemini-2.5-pro-preview-tts	gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts
17 Nov	G	gemini-3.1-flash-live-preview, gemini-2.5-flash-native-audio-preview-12-2025	gemini-3.8-live
30 Nov	A	claude-sonnet-4-5-20250929	claude-sonnet-5-5
30 Nov	O	v1/prompts API and reusable prompt objects	Move prompt content into application code
30 Nov	O	Evals dashboard and API (existing evals read-only from 31 Oct)	Promptfoo (migration guide linked on the page)
30 Nov	O	Agent Builder (ChatKit stays)	Agents SDK or ChatGPT Workspace Agents
1 Dec	O	gpt-image-1-mini, gpt-image-1.5, chatgpt-image-latest	gpt-image-2.5-sunburst or gpt-image-2.5-flare
11 Dec	O	gpt-5-2025-08-07, o3-2025-04-16	gpt-5.6-sol
11 Dec	O	gpt-5-mini-2025-08-07	gpt-5.6-terra
11 Dec	O	gpt-5-nano-2025-08-07	gpt-5.6-luna
11 Dec	O	gpt-5-pro-2025-10-06, o3-pro-2025-06-10	gpt-5.6-sol with reasoning.mode: pro

The OpenAI 23 October entry is long. By family, as OpenAI lists it: gpt-3.5-turbo variants, o4-mini and the fine-tuned ft-gpt-3.5-turbo and ft-o4-mini go to gpt-5.6-terra. gpt-4, gpt-4-turbo, gpt-4-1106-preview, gpt-4o-2024-05-13, o1, o1-pro (with reasoning.mode: pro), o3-mini and ft-gpt-4 go to gpt-5.6-sol. gpt-4.1-nano and its fine-tune go to gpt-5.6-luna. gpt-image-1 goes to the gpt-image-2.5 models. If your code names any gpt-4*, o1* or gpt-3.5* ID, assume it is on this date and read the page for the exact string.

Already past as of today and worth checking in your logs: Google’s antigravity-preview-05-2026 shut down on 5 October (replaced by antigravity-preview-09-2026), and OpenAI’s gpt-5.4-cyber on 1 October and gpt-3.5-turbo-instruct, babbage-002 and davinci-002 on 28 September.

Which dates look like retirements but are not?

Anthropic’s “not sooner than” dates. The status table lists a tentative retirement date for every model, including active ones, and the words are “Not sooner than”. claude-haiku-4-5-20251001 shows 15 October 2026 and claude-opus-4-5-20251101 shows 24 November 2026. Both are listed as Active with no deprecation date. These are floors on when Anthropic could retire them, not announcements. The page says Anthropic provides “at least 60 days’ notice before model retirement for publicly released models”, so the signal to act is a model moving to Deprecated with a retirement date, as Sonnet 4.5 did on 30 September. In Q4 the only Anthropic retirement with a firm date is that one.

Google’s “earliest possible dates”. The Google page says shutdown dates in the table “indicate the earliest possible dates” and that exact dates come with advance notice. Treat Google’s table as a lower bound, and the 15-day gap between the deep research deprecation (8 October) and its shutdown (23 October) as the precedent for how short notice can be.

Routing is not retirement. On 8 October Google deprecated gemini-3.7-flash and gemini-3.5-flash with no shutdown date: “requests to gemini-3.7-flash are automatically routed to gemini-3.8-flash”, and 3.5 Flash to gemini-3.6-flash. Nothing breaks, but you are now running a different model than your tests did. The 2.5 models are a third case: since 18 September access is limited to users who have used them recently, and Google says they are “not deprecated” and served until further notice.

One conflict between Google pages. The Gemini pricing page, as I read it, labelled gemini-2.5-flash-image as shutting down on 2 October 2026, while the deprecations page, updated 9 October, lists 15 March 2027 with gemini-3.1-flash-lite-image as the replacement. I used the deprecations page and left the row out of the table; check both before you rely on it.

One caveat on Anthropic’s dates: they apply to Anthropic-operated platforms (the Claude API, Claude Platform on AWS, Microsoft Foundry). Amazon Bedrock and Google Cloud “set their own retirement schedules”, so a model can live longer there.

What lands just after Q4?

Announced in this quarter, effective later, from the same pages: OpenAI’s gpt-5.3-codex, gpt-5.4-nano and gpt-5.1 retire on 1 April 2027 (to gpt-6-sol or gpt-6-luna), with six months’ notice. OpenAI’s tts-1, tts-1-hd and the 4o mini TTS snapshots go on 6 January 2027, and legacy realtime and audio models on 20 January. Google’s gemini-3.1-flash-lite retires 7 May 2027 (to gemini-3.5-flash-lite).

What should a team do this week?

Three opinions, as design advice, not experience:

1. Grep before you read. Most surprises are model IDs in places nobody remembers: a cron job, a notebook, a fine-tune. Search your repos and your provider’s usage export. Anthropic’s page tells you how: Console, Usage, Export gives usage by API key and model.
2. Order by days left, not by size. The 22 and 23 October entries are 12 and 13 days away. The 30 November Anthropic retirement is 51 days away. A migration with an eval set takes longer than a string change.
3. Pin and flag. Keep model IDs in config, route a slice of traffic to the replacement behind a flag, and compare before the date. Routing changes like Google’s silent 3.7-to-3.8 move are the case where this pays most.

Hands-on: a CI check that fails before the shutdown does

Keep the dates from the table in a file and fail the build when a retiring ID is both present in the code and inside your warning window.

# check_model_ids.py: run in CI. Dates are from the vendors' pages on 10 Oct 2026; re-verify.
import datetime as dt
import pathlib
import re
import sys

RETIRING = {
    "deep-research-pro-preview-12-2025": "2026-10-23",
    "veo-3.1-generate-preview": "2026-10-22",
    "gemini-3.1-flash-live-preview": "2026-11-17",
    "claude-sonnet-4-5-20250929": "2026-11-30",
    "gpt-5-2025-08-07": "2026-12-11",
    "o3-2025-04-16": "2026-12-11",
}
WARN_DAYS = 60
today = dt.date.today()
failed = False

for path in pathlib.Path("src").rglob("*"):
    if not path.is_file() or path.suffix not in {".py", ".ts", ".js", ".json", ".yaml", ".yml", ".md"}:
        continue
    text = path.read_text(errors="ignore")
    for model_id, date in RETIRING.items():
        if re.search(re.escape(model_id), text):
            days = (dt.date.fromisoformat(date) - today).days
            if days <= WARN_DAYS:
                print(f"{path}: {model_id} retires {date} ({days} days)")
                failed = True

sys.exit(1 if failed else 0)

The table is deliberately short: add the IDs you actually use, and add the Anthropic floors only after they become real deprecations.

Takeaways

• Q4 has dated retirements on 22 and 23 October, 17 and 30 November, 1 and 11 December. The hard Anthropic one is claude-sonnet-4-5-20250929 on 30 November, replacement claude-sonnet-5-5.
• Anthropic’s “not sooner than” dates and Google’s “earliest possible dates” are floors, not announcements. Act on a Deprecated status with a date.
• Silent routing, as with Gemini 3.7 Flash to 3.8 Flash, changes behaviour without an error. Test for it.
• OpenAI’s 23 October list covers most gpt-3.5, gpt-4, o1 and o4-mini IDs, and the pages show your migration targets are the gpt-5.6 family.
• Put model IDs in config and check them in CI against a dated list.

Next, the Gemini deep research post covers the 23 October Google shutdown in detail.

Sources

1. Claude docs: model deprecations (status table, history, notice policy) · platform.claude.com
2. Gemini API deprecations (shutdown table, page updated 9 October 2026) · ai.google.dev
3. Gemini API pricing (listed gemini-2.5-flash-image differently from the deprecations page) · ai.google.dev
4. Gemini API release notes (August to October 2026 entries) · ai.google.dev
5. OpenAI API docs: deprecations (the platform.openai.com URL redirects here) · developers.openai.com