When an email with "deprecation" or "retire" in the subject line arrives, my hand stops on the mouse before I open it.
I'm an indie developer who calls the Claude API from my own app backends, and it still happens. If Claude runs behind a no-code connector, or behind something a contractor built for you, the English text may not say what you need to do or by when. Closing the email and telling yourself "later" is a very human reaction, and I did exactly that at first.
Here is the first thing I'd like you to know: you don't need to change anything on the day you read it. You need to open three screens and count three facts.
The breathing room the docs promise
On October 27, 2026, I checked the Sonnet 4.5 (claude-sonnet-4-5-20250929) entry on the official model deprecations page.
| Item | What the page says |
|---|---|
| Status | Deprecated, announced September 30, 2026 |
| Retirement date | November 30, 2026 (marked as tentative) |
| Recommended replacement | claude-sonnet-5-5 |
| After retirement | Requests to that model fail |
The same page says that for publicly released models, customers with active deployments are notified by email and in the documentation at least 60 days before retirement. From September 30 to November 30 is 61 days. Counting from today, 34 days remain.
That is enough time. It is also exactly the length of time that disappears if you say "later" three times.
Screen one: does this email concern you?
Start with who the notice is for, not what it says.
It goes to accounts that call Claude through the API. If you only open the Claude chat window and talk, you never specify a model ID, and the official page is written about the API, so you can relax here.
It may concern you if any of these is true:
- You once logged in to the Console and created an API key
- You registered an Anthropic API key in a no-code connector
- A contractor or friend built something on Claude for you
The third case is the one people miss. If any of them sounds familiar, open the Usage page in the Console and look at recent usage by model. The Export button gives you a CSV broken down by API key and model, which any spreadsheet can open.
If no row shows up at all, the notice may be about an old key. Write "no action needed" somewhere you will find it, and you are done.
Screen two: read the date as a deadline
The November 30 date carries a "tentative" note. Tentative means it can move, and it may move earlier. I read it as "by this day", never as "probably later".
After that date, requests naming the model fail. The page does not spell out the exact error text, so in practice you would notice because failures start piling up in a connector's run history.
What I put on my calendar is not the retirement date but the day two weeks before it: November 16. If everything is checked by then, the real date no longer makes my stomach hurt.
Screen three: who do you tell, and what?
If you found rows in Usage, the next question is who will fix them.
If it's you, the model name in your code or settings looks like this:
{
"model": "claude-sonnet-4-5-20250929",
"max_tokens": 1024,
"messages": [{"role": "user", "content": "Hello"}]
}The model line is what gets replaced. The recommended replacement is claude-sonnet-5-5, so it ends up as:
{
"model": "claude-sonnet-5-5",
"max_tokens": 1024,
"messages": [{"role": "user", "content": "Hello"}]
}Swapping the name alone doesn't always work, though. Models can differ in which settings they accept, and leftover old settings may be rejected. I've written about how to check that in other articles:
- 43 Days Until Sonnet 4.5 Retires: I Counted Callers From the Usage CSV, Not From grep
- claude-sonnet-4-5 stops on November 30 — I checked the bill before calling its replacement cheaper
If someone else maintains it, three lines are enough to send them:
- Anthropic sent a retirement notice for
claude-sonnet-4-5-20250929 - Retirement is November 30, 2026 (tentative), and the recommended replacement is
claude-sonnet-5-5 - I'd like testing finished by November 16
Forwarding the email works too, but adding these three lines tells the other person where to start.
Count first, rename last
The one rule I keep every time is the order: count first, rename last.
Rename in a hurry and you can break things that were working. If you count in Usage and find the affected places before touching anything, you can replace them one at a time, and if something fails you can trace why.
Today, I'd suggest opening Usage in the Console and checking only whether claude-sonnet-4-5 appears in the last 30 days of per-model rows. That alone will change how heavy this email feels.