Claude Sonnet 4.5 retires on November 30: every date, what breaks and how to switch

Claude Sonnet 4.5 retires from the Claude API on November 30, 2026, according to Anthropic’s model deprecations page. After that date, requests to claude-sonnet-4-5-20250929, or its short alias claude-sonnet-4-5, fail. Anthropic deprecated the model on September 30, emailed the developers who use it, and recommends claude-sonnet-5-5 as the replacement.

The date isn’t the same everywhere: Amazon Bedrock keeps Sonnet 4.5 running until April 2027, and Google Cloud hasn’t set one at all. Switching is also more than a one-line change, because Sonnet 5.5 rejects several settings that Sonnet 4.5 accepted. Here is every date, what breaks and what to do.

Sonnet 4.5 deprecation date and retirement date

Anthropic released Claude Sonnet 4.5 on September 29, 2025. A year and a day later it moved the model to “deprecated”, which Anthropic defines as “still functional but no longer recommended”, with a fixed retirement date. After that, “requests to retired models will fail.”

The November 30 date covers the platforms Anthropic runs itself: the Claude API, Claude Platform on AWS and Microsoft Foundry. These are the IDs to search your code for:

From Anthropic’s Sonnet 4.5 model page and deprecations page, AWS’s Bedrock model lifecycle page and Google Cloud’s Sonnet 4.5 page.

Anthropic also warns that “deprecated models are likely to be less reliable than active models”, which is a good reason to move well before the deadline rather than on November 29.

Sonnet 4.5 on Bedrock and Google Cloud

Amazon and Google set their own schedules for Claude, and AWS is blunt about it: “For Amazon Bedrock usage, only the dates on this page apply.” Its model lifecycle page moved Claude Sonnet 4.5 to “Legacy” on October 8, 2026, which sets three stages:

• Now: new customers can’t start using it, existing customers “may lose access” after 15 days of inactivity, and you can’t buy new Provisioned Throughput for it.
• January 8, 2027: “public extended access” begins. The model keeps working, but AWS says to “expect higher pricing, which will be set by the model provider”.
• April 8, 2027: end of life, after which requests fail.

Those Bedrock dates apply in 31 AWS regions worldwide, including all four US regions, Canada, Mexico, São Paulo, Ireland, London, Frankfurt, Paris, Stockholm, Zurich, Milan, Spain, Tel Aviv, Cape Town, Mumbai, Tokyo, Seoul, Singapore, Sydney and Jakarta. AWS notes that lifecycle states can differ by region, so check the console for the one you use.

Google Cloud still lists Claude Sonnet 4.5 as generally available on its Sonnet 4.5 page, with only a “retirement date not sooner than” September 29, 2026, and the model isn’t on Google’s partner model deprecations list. Google has given earlier Claude models six months between deprecation and shutdown: Claude 3.7 Sonnet, for example, was deprecated on November 11, 2025, and shut down on May 11, 2026.

Every Claude model retirement date

Sonnet 4.5 is the only Claude model with a set retirement date right now. For active models, Anthropic lists a “not sooner than” date, the earliest it could retire them rather than a plan, and it promises customers with active deployments “at least 60 days’ notice” before a publicly released model goes.

From Anthropic’s model deprecations page. The dates apply to the Claude API, Claude Platform on AWS and Microsoft Foundry.

The one to watch is Claude Haiku 4.5. Its floor is October 15, 2026, a week after Anthropic launched its replacement, Claude Haiku 5.5, at a tenth of the price. Opus 4.5’s floor follows on November 24.

Already retired on the Claude API: Claude Opus 4.1 (August 5, 2026), Claude Opus 4 and Sonnet 4 (June 15, 2026), Claude Haiku 3 (April 20, 2026), Claude Haiku 3.5 and Sonnet 3.7 (February 19, 2026) and Claude Opus 3 (January 5, 2026). Bedrock runs later: Claude Sonnet 4 reaches end of life there on October 14, 2026, and Claude Opus 4.1 on January 8, 2027.

Sonnet 4.5 to Sonnet 5.5: what breaks

Changing the model name is only step one. Anthropic’s Sonnet 5.5 migration guide lists several settings that Sonnet 4.5 accepted and Sonnet 5.5 rejects with a 400 error. If your app fails after the swap, check these first:

• Prefill. Ending with a half-written assistant message, a common way to force JSON, returns This model does not support assistant message prefill. The conversation must end with a user message. Use structured outputs instead, or move the instruction into the system prompt or user message.
• Thinking budgets. "thinking": {"type": "enabled", "budget_tokens": N} returns "thinking.type.enabled" is not supported for this model. Remove it and set an effort level. Anthropic says there’s no fixed mapping from a budget to an effort level, so test two or three.
• Temperature, top_p and top_k. Any non-default value returns an error on Claude 4.7 and later models. Anthropic’s Python SDK drops the three parameters from v1.0, so passing them raises a TypeError before the request is even sent.
• Forced tool use. A tool_choice of any or tool fails with tool_choice: type "tool" and "any" are not supported for this model. Send auto with a strict tool instead. On Bedrock, where strict tools aren’t available for Sonnet 5.5, send auto and check the tool input in your own code.
• Old parameters and headers. Move output_format to output_config.format, remove the interleaved-thinking-2025-05-14 and context-window beta headers, and swap fine-grained-tool-streaming-2025-05-14 for eager_input_streaming. Computer use needs a newer tool version than computer_20250124.

The quieter change is that Sonnet 5.5 thinks by default, even when a request has no thinking field. Replies can start with a thinking block, so code that reads content[0].text breaks, and thinking tokens count towards max_tokens and are billed as output. To switch up-front thinking off, send {"type": "between_tools"}, because "disabled" also returns an error.

Is Sonnet 5.5 cheaper than Sonnet 4.5?

US dollar prices from Anthropic’s pricing page and model pages.

Per token, Sonnet 5.5 is a third cheaper. But it uses a newer tokeniser that turns the same text into about 30% more tokens, Anthropic says, and images cost more: a 2000×1500 picture uses about 2.5 times as many tokens. Plain text still comes out slightly cheaper, but default thinking can wipe out the saving, so compare a few days of real bills. Our Sonnet 5.5 launch story covers how it scores against Opus 5.5.

What to do if you got the Sonnet 4.5 email

1. Find where you use it. In the Claude Console, open the Usage page, click Export and check the CSV, which breaks usage down by API key and model. Then search your code, config files and environment variables for claude-sonnet-4-5, which catches both the dated ID and the alias.
2. Let Claude Code do the first pass. Run /claude-api migrate this project to claude-sonnet-5-5 in Claude Code. Anthropic’s bundled Claude API skill swaps the model ID and the breaking settings, asks you to confirm the scope first, and leaves a checklist of things to check by hand.
3. Test on real work. Run your usual prompts at two or three effort levels and compare answers, speed and cost. Max and Team subscribers can use their free monthly API credits for this.
4. Check Claude Code’s model settings. On Amazon Bedrock, Google Cloud and Microsoft Foundry, Claude Code’s sonnet alias still points to Sonnet 4.5, its model configuration docs say. Foundry follows Anthropic’s November 30 date, so set ANTHROPIC_DEFAULT_SONNET_MODEL to your Sonnet 5.5 model or deployment name. Claude Code shows a warning at start-up when your chosen model has a retirement date.
5. Ask your suppliers which Claude model their tools call, and when they are switching.

Does it affect the Claude app or GitHub Copilot?

The deprecation notice is about the API. If you only chat on claude.ai or in the Claude apps, there is no code to change: you pick from whatever the model menu offers, with older models under “More models”, and Anthropic hasn’t published a separate app date for Sonnet 4.5.

GitHub runs its own timetable and has already moved on: Sonnet 4.5 left GitHub Copilot on September 1, 2026, as our guide to Claude missing from Copilot explains. OpenAI is clearing out its API too, with a batch of older models shutting down on October 23.

Will Sonnet 4.5 come back?

Anthropic hasn’t said it will bring Sonnet 4.5 back after November 30. It has committed to keeping the weights of every publicly released model for at least the lifetime of the company, so a return isn’t ruled out.

So far it has made one exception: Claude Opus 3, retired on January 5, 2026, is still available to paid claude.ai users and on the API by request. But Anthropic said it is “not committing to similar actions for every model in the future”, because the cost of serving old models “scales roughly linearly” with each one.

Last checked: October 13, 2026, against Anthropic’s model deprecations page, AWS’s Bedrock model lifecycle page and Google Cloud’s Claude pages.

Frequently asked questions

When is Claude Sonnet 4.5 being retired?

Claude Sonnet 4.5 retires from the Claude API on November 30, 2026. After that, requests to claude-sonnet-4-5-20250929 or the claude-sonnet-4-5 alias fail. The same date applies on Claude Platform on AWS and Microsoft Foundry.

Is Sonnet 4.5 deprecated?

Yes. Anthropic deprecated Claude Sonnet 4.5 on September 30, 2026, and emailed developers who use it. Deprecated means it still works but is no longer recommended, and it has a fixed retirement date of November 30, 2026.

Is Sonnet 4.5 still available?

Yes, for now. It works on the Claude API until November 30, 2026. On Amazon Bedrock existing customers can use it until April 8, 2027, with higher prices possible from January 8, 2027. Google Cloud hasn’t set a shutdown date.

What replaces Claude Sonnet 4.5?

Anthropic recommends Claude Sonnet 5.5, model ID claude-sonnet-5-5. It costs $2 per million input tokens and $10 per million output tokens, against $3 and $15 for Sonnet 4.5, and has a 1 million token context window.

When is Sonnet 4.5 deprecated on Bedrock?

Amazon Bedrock moved Claude Sonnet 4.5 to Legacy on October 8, 2026. Public extended access, where AWS says to expect higher pricing, starts on January 8, 2027, and the model reaches end of life on April 8, 2027.

Is Sonnet 5.5 cheaper than Sonnet 4.5?

Per token, yes: about a third cheaper. But Sonnet 5.5 turns the same text into about 30% more tokens and thinks by default, with thinking billed as output, so compare real bills before assuming a saving.

Why does my code return a 400 error on Sonnet 5.5?

Usually one of four settings that Sonnet 4.5 accepted: an assistant prefill, a thinking budget (thinking type enabled), a non-default temperature, top_p or top_k, or a tool_choice of any or tool. Anthropic’s Sonnet 5.5 migration guide lists the replacement for each.

Is temperature deprecated in the Claude API?

Yes, on Claude 4.7 and later models, including Sonnet 5.5. A non-default temperature, top_p or top_k returns a 400 error, and Anthropic’s Python SDK from v1.0 removes the parameters, so passing them raises a TypeError.

Will Sonnet 4.5 come back?

Anthropic hasn’t said so. It keeps the weights of every publicly released model, and it kept Claude Opus 3 available to paid claude.ai users after retirement, but it says it isn’t committing to do the same for every model.

Does the Sonnet 4.5 retirement affect the Claude app?

The notice is about the API. In the Claude app you choose from whatever the model menu offers, and Anthropic hasn’t published a separate app date for Sonnet 4.5. There is no code to change if you only use the app.

Sources: Anthropic (model deprecations), Anthropic (Claude Sonnet 4.5), Anthropic (Sonnet 5.5 migration guide), Anthropic (pricing), Anthropic (Claude Code model configuration), Amazon Web Services (Bedrock model lifecycle), Google Cloud (Claude Sonnet 4.5).