Claude Sonnet 4.5 retires on November 30, 2026: what stops working and what you can save

Updated November 2, 2026

Anthropic retires Claude Sonnet 4.5 from the Claude API on November 30, 2026. After that, requests to claude-sonnet-4-5-20250929 fail, and Anthropic names Sonnet 5.5 as the replacement.

In claude.ai, a help-center note archived May 17, 2026 said it would leave the model selector on May 15. If you still reach it through the API or an app built on it, you have until the date above. Neither Anthropic page gives a time of day, so plan to finish before November 30 starts.

Either way, you can save your chats, your memory and your instructions. The model itself can’t be saved.

We make Counterparts, a memory for Claude, so we have a stake in the last section; everything before it is useful without it. An AI drafted this page from the sources at the foot, read again on November 2, 2026, and another AI checked it. No person has run the steps or the prompt in step 4.

What happens on November 30

Where you use Sonnet 4.5	What we know
Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed), Microsoft Foundry	Requests fail after the retirement date.
Amazon Bedrock	AWS lists an end-of-life date of April 8, 2027, and the state can differ by Region. See the Bedrock question below for the catches.
Claude Models on Google Cloud (Google managed)	Google sets its own schedule. Check its model table.
claude.ai and the Claude apps	November 30 is the API date. A help-center note archived May 17, 2026 said Sonnet 4.5 would be removed from the model selector on May 15; later copies don’t have it, and we found no confirmation of the removal. Opus 4 and 4.1 also left the selector months before their API retirements.
Your existing chats	Anthropic hasn’t said for Sonnet 4.5. When Haiku 4.5 left the selector, Anthropic said its chats stayed and continued on Haiku 5.5 (see the question below). Export now anyway.

Other Claude and ChatGPT models’ dates are in our retirement dates guide.

What to save, in order

Export your chats. On the web or in Claude Desktop, go to Settings > Privacy and click Export data; you can’t start this from the phone apps. A link arrives by email and expires after 24 hours. On a Team or Enterprise plan, an owner has to allow it, and the export leaves out the contents of files you uploaded. If you talk to Sonnet 4.5 through another app, save your chats from that app too.
Save Claude’s memory of you. In Settings > Your Claude, select Manage next to Your memory (Settings > Memory in the older layout), or ask Claude to “Write out your memories of me verbatim, exactly as they appear in your memory.” Keep that text in a file.
Copy your project instructions and styles into a file, so you have them when you set up the new model.
Write down what you liked. Anthropic suggests having Claude review conversations you enjoyed and write instructions from them. Paste a few conversations in and use the prompt below, with Sonnet 4.5 if you can still reach it, or whichever model you have now.
Try Sonnet 5.5 now, side by side, with your own prompts. Anthropic recommends testing before the old model goes, and you’ll have time to find out what you need to adjust.

The prompt for step 4:

Read the conversations I've pasted below and write instructions I could give a new Claude model so that it works with me the way I liked working with the model in them. Cover tone, length, how it handled disagreement, what it asked or looked up before answering where you can see that in the text, and things I corrected it on. Quote examples only from what you can see here, and if you can't find an example for something, say so and don't make one up. Tell me which instructions you're least sure of. Write it so I can paste it into a project or into my account's instructions.

Treat the result as a list of what you liked. A new model will follow it in its own way.

If you build on the API

Find what still calls it: in Claude Console, open Usage and click Export for a CSV by API key and model. Check anywhere the model ID is pinned, including Claude Code settings.
Don’t only swap the model name. Among the things Sonnet 5.5 rejects with a 400 error: temperature, top_p or top_k; a thinking budget (budget_tokens); a prefilled last assistant turn; a forced tool_choice (any or tool); thinking: {"type": "disabled"} (send between_tools instead); output_format (move it to output_config.format); and the computer_20250124 tool. It also thinks on requests that don’t mention thinking, so a response can start with a thinking block, and code that reads content[0].text breaks: read blocks by type. The same text counts about 30 percent more tokens. Anthropic’s Sonnet 5.5 migration guide has the full checklist by starting model.
If you need a closer step, Sonnet 4.6 is still active, with a date of “not sooner than February 17, 2027”. That’s a floor, and it will retire too.

What carries over to Sonnet 5.5

Carries over	Doesn’t
Claude’s memory, if you have it turned on	The way Sonnet 4.5 wrote
Instructions you paste in, or that sit on your account	A conversation’s feel, even if the text is saved

If Sonnet 4.5 was your selected model, expect to have been moved to a newer one: when Haiku 4.5 left, Anthropic switched people to Haiku 5.5. We found nothing on projects, so check the model next to the send button in any project you use.

If the new model feels too chatty, too terse or different in tone, Anthropic says you can adjust that with instructions. That helps, but Sonnet 5.5 is a different model and won’t write like Sonnet 4.5.

Your export is an archive you keep: Anthropic says you can’t import one into another personal account.

Could Sonnet 4.5 come back?

Don’t plan on it. Anthropic has committed to keeping the weights of its publicly released models for at least the life of the company, and says it hopes to make past models available again. It hasn’t promised to.

There’s one precedent. Claude Opus 3 retired on January 5, 2026, and as of February 2026 paid claude.ai subscribers could still use it, with API access by request. Anthropic also wrote that it’s “not committing to similar actions for every model.”

Where Counterparts fits

Start with Claude’s built-in memory. It’s already in your account, and Anthropic recommends it for helping a new model pick up your preferences. Anthropic also says it focuses on work-related topics, and that imported personal details unrelated to work may not be kept.

Counterparts keeps one memory of the whole relationship: you, your work together, and what the AI learned about itself. The AI writes it down in its own words, in a store that sits outside any one model. Any Claude model you use with it starts each session from a short note on who you are and where you left off, and related memories come up as you talk. A memory written with Sonnet 4.5 this month can come up for Sonnet 5.5 in December, though memories fade unless they come up again. Because the memory sits outside any one model, it carries on through the next retirement too. Sonnet 5.5 reads the notes, but it isn’t the one who wrote them, and Counterparts doesn’t change the model. It doesn’t read your chat export or save transcripts; it starts remembering from your first conversation with it.

It runs today in Claude Code on Mac and Linux, and in Claude Desktop on macOS, where memories stay on your machine; that install is free and open source. A version for claude.ai is in beta and needs an invite for now; to ask for one, email mike@resonantventures.com. It stores memories on our server. Install it for Claude Code, or see the Claude Desktop steps.

Questions people ask

Will Sonnet 5.5 know my preferences?

If you’ve turned on Claude’s memory, Anthropic says it can help a new model pick them up. The memory is a set of notes, so the new model will know things about you without having been the one who talked to you.

What happens to my Sonnet 4.5 chats?

Anthropic hasn’t said for Sonnet 4.5. When Haiku 4.5 left the model selector for Free, Pro and Max on October 29, 2026, Anthropic said its chats were still in the account, nothing was deleted, and continuing one got a reply from Haiku 5.5. Export yours anyway, so you have a copy either way.

Can I keep using Sonnet 4.5 through Bedrock?

If you already use it there, yes, for now. AWS lists Sonnet 4.5 as Legacy since October 8, 2026, with an end-of-life date of April 8, 2027. New customers can’t use Legacy models, and existing customers may lose access after 15 days of inactivity. From January 8, 2027 it enters public extended access, where AWS says to expect higher pricing.

Sonnet 4.5 isn’t in my picker. What can I still do?

Steps 1 to 4 above still work with whichever model you have now. If you want Sonnet 4.5 itself for step 4, the API still serves it until November 30.

Is Opus 4.5 next?

Its API date is “not sooner than November 24, 2026”, which is a floor and not an announcement. Anthropic gives at least 60 days’ notice, and Opus 4.5 had none as of November 2, 2026. If a notice went out that day, the earliest API retirement would be January 1, 2027 (our arithmetic from the 60-day rule). The claude.ai picker runs on its own schedule, so check “More models” too, and the deprecations page for a notice since.

Questions? Ask in r/AIMemorySystems.

Sources

All read on November 2, 2026. If something here has gone stale, tell us.