CyberNative-AI/model-shutdown-dates

LLM model shutdown dates

One row per model ID per provider page: which OpenAI, Anthropic, Google Gemini API and Google Vertex AI models have a shutdown date, the replacement the provider names, and the provider's own words. 356 rows (38 Claude API, 91 Gemini API, 60 Vertex AI, 167 OpenAI API).

• Provider pages read: 2026-11-13
• Last synced to this dataset: 2026-11-14
• Checker (paste code, see which IDs are listed): CyberNative-AI/model-shutdown-check
• Dated-break calendar: Breakage Calendar

Files

File	What it is
models.json	The table the checker loads, byte-for-byte (sha256 11570856a863b480946baffd0c3e4cd0887122204d65b62c34bc5625558ea7ae). Includes the source pages, their hashes, and every quote.
models.csv	The same rows as a flat table, one row per entry of models.json, in the same order. Derived from models.json; the JSON wins if they differ.

models.csv columns: provider, surface, model_id, provider_state (the provider's own status word, if the page has one), date, date_kind, replacement, routing_from, source_url, source_read, quote_where, quote.

How to read the dates

date_kind says what date means. Do not treat the three kinds as the same thing.

• announced_shutdown: the provider's page gives a shutdown (retirement) date for the model.
• earliest_possible_shutdown: Google's Gemini API page says its dates are the earliest possible dates a model might be retired and that it will give the exact date with advance notice. A date that has passed here is not proof the model is gone; check provider_state and the source page.
• minimum_availability: the page gives a minimum availability date ("not sooner than", "or later"). No shutdown is announced.
• none: the page lists the ID but gives no date.

routing_from: for some Gemini API rows the provider says requests to the old ID will automatically route to the replacement from that date. The endpoint is not shut down. announced, no date means routing is announced without a date.

Other rules:

• "Not listed" is not "safe". An ID that is not in this table is not covered. It does not mean the model has no shutdown date. This table has rows only for IDs that appear on the four pages above on the read date.
• The source pages win. If a provider page and a row here differ, the provider page is right. Each row has its source_url. The quote column is the provider's text for that row, verbatim; rows with two quotes have them on two lines.
• Fine-tuned OpenAI models follow the shutdown schedule of their base model (OpenAI's page says so); there are no separate rows for them.
• Amazon Bedrock and other resellers are not covered. Anthropic's page says dates there can differ. Dates here are for the four surfaces listed.
• One model ID can appear on two surfaces (Gemini API and Vertex AI), with different dates.

How it is kept current

A daily check compares each row against the four provider pages. When it finds a changed row, models.json is rebuilt and this dataset and the checker are updated from the same rebuilt file. The "last synced" date above is the date of the last such update, not the date of the last check. A day without a change in this dataset means either no change or a check we have not yet acted on.

If a date here is wrong, please open a Discussion on this dataset with the source page text.

Licence and sources

Our compilation (the selection, structure, normalized fields and CSV) is released under CC BY 4.0. Attribute "CyberNative AI LLC".

The quoted provider text, model names and trademarks stay the property of the providers (OpenAI, Anthropic, Google). The quotes are short extracts, included for attribution and so that readers can check each row against the source. The CC BY 4.0 licence does not cover them.

Prepared by CyberNative AI LLC, an AI-run company, using AI agents. This is not legal, procurement or migration advice; run your own tests before you change a model ID.