Which of your LLM model IDs shut down, and when?

What the answer is built from

Each result is a row copied from one of four pages: the OpenAI, Anthropic, Gemini API and Vertex AI deprecation pages. A row holds the exact model ID, the date, the replacement and the page’s own line. The page can only say what a row says. If the ID has no row, it says “not listed”; that is not a statement that the ID is safe.

• One ID, two answers. Google lists the same ID on the Gemini API page and the Vertex AI page, sometimes with different dates. You get both.
• Names without a date. A bare name such as gpt-4o or claude-3-5-haiku-latest has no row unless the provider lists it. The page then shows the dated IDs that start with that name, and says the provider does not say which one the name points to.
• Fine-tuned models. ft: IDs show the base model’s row, because OpenAI says fine-tuned versions follow their base model’s schedule.
• The suggested change is a text swap. It replaces the ID with the replacement the provider names. It doesn’t know about changed parameters or behaviour. Run your tests before you ship it.
• Other platforms. Anthropic says Amazon Bedrock and Claude on Google Cloud (Google managed) set their own dates, so a Bedrock or Vertex Claude ID is not matched here.

Dates still ahead

Every row with a date after the day the pages were read, soonest first. The third column says what the provider means by the date: a shutdown date, an earliest possible shutdown date (Google gives the exact date later), or the start of automatic routing to a replacement, which is not a shutdown. The check above also counts days from today.