AI model retirement calendar: OpenAI, Anthropic and Google through April 2027

1. The calendar#

All three providers publish retirement dates and the recommended replacement on their deprecation pages. The table gathers them ordered by date, as read on October 19, 2026. After the retirement date, requests to a retired model fail.

Google's dates are the earliest possible dates on which a model might be retired: the page itself says exact dates will be communicated in advance, so they may change. Anthropic says it gives at least 60 days' notice before retiring a public model.

Documentation: OpenAI: deprecations ↗ · Anthropic: model deprecations ↗ · Google: Gemini API deprecations ↗

2. How many days you have left#

The script stores the calendar in a list, computes the days remaining from a reference date (here October 19, 2026) and flags with an exclamation mark what falls within the next 45 days. Change the reference date to today's and add your own models.

from datetime import date

TODAY = date(2026, 10, 19)
WARN_DAYS = 45

RETIREMENTS = [   # (date, provider, models, recommended replacement)
    (date(2026, 10, 22), "Google", "gemini-omni-flash-preview, veo-3.1-*-preview", "gemini-omni-1.1-flash"),
    (date(2026, 10, 23), "Google", "deep-research-pro-preview-12-2025", "deep-research-preview-04-2026"),
    (date(2026, 10, 23), "OpenAI", "gpt-4-turbo, gpt-4.1-nano, o1, o3-mini, o4-mini, gpt-image-1", "gpt-5.6-sol/terra/luna"),
    (date(2026, 11, 17), "Google", "TTS and Live previews (3.1 and 2.5)", "gemini-3.8-flash-tts, gemini-3.8-live"),
    (date(2026, 11, 30), "Anthropic", "claude-sonnet-4-5-20250929", "claude-sonnet-5-5"),
    (date(2026, 11, 30), "OpenAI", "Evals platform", "Promptfoo"),
    (date(2027, 1, 6), "OpenAI", "tts-1, tts-1-hd, gpt-4o-mini-tts", "gpt-realtime-2.1-mini"),
    (date(2027, 1, 14), "OpenAI", "gpt-4o, gpt-4o-mini, gpt-4.1, gpt-4.1-mini", "gpt-6-luna"),
    (date(2027, 4, 1), "OpenAI", "gpt-5.3-codex, gpt-5.1, gpt-5.4-nano", "gpt-6-sol / gpt-6-luna"),
]

def upcoming(provider=None):
    return sorted(r for r in RETIREMENTS if provider in (None, r[1]) and r[0] >= TODAY)

urgent = 0
for when, provider, models, replacement in upcoming():
    days = (when - TODAY).days
    urgent += days <= WARN_DAYS
    flag = "!" if days <= WARN_DAYS else " "
    print(f"{flag} {when} | {days:>3} d | {provider}: {models} -> {replacement}")
print(f"{urgent} retirements in the next {WARN_DAYS} days")

A warning from the source: OpenAI's October 23 retirements recommend replacements from the gpt-5.6 family, while those of April 2027 point to gpt-6-sol and gpt-6-luna. Check OpenAI's page to pick the current replacement for your case.

3. Anthropic: beyond November 30#

Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) has been deprecated since September 30, 2026 and retires on November 30, with claude-sonnet-5-5 as the replacement. Other models have tentative «not sooner than» dates: claude-opus-4-5-20251101, not sooner than November 24, 2026, and claude-haiku-4-5-20251001, not sooner than October 15, 2026, both still active. A «not sooner than» date is not a retirement date.

When migrating, keep a parameter change in mind: on Claude 4.7 and later models, sending temperature, top_p or top_k with a non-default value returns a 400 error. Anthropic's recommendation is to omit them and guide behavior with the prompt. Python SDK 1.0 and later no longer includes those parameters.

4. How to audit your usage#

• Anthropic lets you export from the console a CSV of usage by API key and model, to see which retirable models you use.
• Search your code for hardcoded model identifiers (strings like gpt-4o or claude-sonnet-4-5) and your environment variables.
• Also check the third-party services that call those models for you.
• Test the replacement with your own evaluations before the retirement date; replacements are not identical.
• If you use OpenAI's Evals platform, it becomes read-only on October 31 and shuts down on November 30.

5. A simple migration plan#

• Inventory models by provider, with who uses them and for what.
• Sort by retirement date and start with what falls first.
• Pick the recommended replacement and compare costs with Codifly's AI model comparator.
• Run your evaluations on both models and decide with data.
• Deploy behind a flag you can roll back and retire the old model after one stable week.

Go deeperHow to Choose an LLM for Production Without Relying on Leaderboards

6. Keep your numbers current#

The dates are those each provider published as of October 19, 2026. They change often; reread the three pages before acting and note the reading date.

• Subscribe to each provider's deprecation notices.
• Automate an alert when fewer than 60 days remain for a retirement that affects you.
• Keep your evaluations ready to test a replacement in a day.