Claude Sonnet 4.5 Retirement: What the November 30 Date Means

Claude Sonnet 4.5 is being retired. Anthropic marked the model deprecated on the Claude API on September 30, 2026, and has set a tentative retirement date of November 30, 2026, after which requests to claude-sonnet-4-5-20250929 will fail on Anthropic's own platforms. The recommended replacement is claude-sonnet-5-5. Here's what the dates actually mean, who is affected, and how to move your workloads before the cutoff.

What's actually happening to Claude Sonnet 4.5

Anthropic publishes a running model-lifecycle page that tracks every Claude model through four stages: Active (fully supported), Legacy (no further updates, may be deprecated later), Deprecated (still functional, no longer recommended, with a replacement and a retirement date attached), and Retired (no longer available; requests fail outright). On September 30, 2026, Anthropic moved Claude Sonnet 4.5 from Active into the Deprecated column and attached a retirement date to it.

That means Sonnet 4.5 keeps answering requests normally right now. What changes is the clock: Anthropic's documentation now lists a tentative retirement date, and once that date arrives, calls to the model will stop working rather than degrade gracefully.

The November 30, 2026 date, and why "tentative" matters

Anthropic's own model-status table lists the retirement date for claude-sonnet-4-5-20250929 as November 30, 2026 — but every forward-looking date on that page, including this one, is labeled a tentative retirement date, not a guaranteed one. For models that haven't yet been deprecated, Anthropic phrases the column as "not sooner than" a given date, which is a floor rather than a fixed day. Once a model is actually deprecated, as Sonnet 4.5 now is, the tentative date functions as the real target, but Anthropic still reserves language that treats it as subject to change rather than contractually locked.

In practical terms: plan around November 30, 2026, but don't assume the exact calendar day is etched in stone the way a product launch date would be. Anthropic says it notifies customers with active deployments at least 60 days before a model is actually retired, which — counting back from November 30 — lines up closely with the September 30 deprecation notice itself.

Who this actually affects

This is primarily a developer and enterprise story, not a change to the Claude.ai chat app's model picker for most everyday users. The retirement dates on Anthropic's deprecations page apply specifically to:

• The Claude API (direct API access via api.anthropic.com)
• Claude Platform on AWS
• Claude Models on Google Cloud, in its Anthropic-managed configuration
• Microsoft Foundry

If your product calls Claude Sonnet 4.5 directly through any of those four surfaces, you have until the retirement date to migrate. Two platforms are explicitly carved out: Amazon Bedrock and the Google-managed configuration of Claude on Google Cloud each set their own retirement schedules, which can run on a different timeline than Anthropic's own platforms. If you're running Sonnet 4.5 through Bedrock, check Bedrock's own model lifecycle table rather than assuming the November 30 date applies.

Consumers are affected indirectly: any third-party app, chatbot, or internal tool built on top of Sonnet 4.5 through the API will stop working past the retirement date unless its developers update it, so "the app I use suddenly broke" complaints are a realistic downstream effect even for people who've never touched the API themselves. The same applies to internal company tools — a support bot, a document summarizer, or a coding assistant that an engineering team wired directly to claude-sonnet-4-5-20250929 months ago is just as exposed as a public-facing product, and these are the integrations most likely to get missed because nobody owns them full-time.

Enterprise customers running Sonnet 4.5 at scale have an added wrinkle: Anthropic's 60-day notice policy is a floor, not a guarantee of hand-holding. The company says it will notify "customers with active deployments," which assumes your account is reachable at the email on file and that whoever receives it understands it applies to production traffic. Teams that inherited an integration from a departed engineer, or that route API keys through a shared service account, are the ones most likely to discover the retirement the hard way — when requests start failing — rather than from the notice itself.

Sonnet 4.5's predecessor, Claude Sonnet 4, went through the exact same cycle earlier this year — deprecated April 14, retired June 15 — so the roughly two-month deprecation-to-retirement window is consistent with how Anthropic has handled the last several Sonnet and Opus generations.

Why Anthropic is retiring it now

Anthropic's documentation is direct about the reason: it deprecates and retires older models "to ensure capacity for new model releases." Claude Sonnet 5.5 and Claude Opus 5.5 are both now Active, and Anthropic's model table shows Sonnet 5.5 isn't due for its own retirement consideration until "not sooner than September 28, 2027" — more than a year of runway, which is the kind of gap Anthropic generally leaves before the next deprecation cycle starts.

Anthropic also acknowledges the retirement isn't free of downsides. Its documentation lists three costs directly: users who prefer a specific model's behavior or "character" are forced onto a newer one even if they liked the old one better; researchers running longitudinal comparisons lose access to a fixed baseline; and model retirement carries what Anthropic calls "safety- and model welfare-related risks" — company language that reflects internal alignment research finding that some Claude models have, in testing scenarios, resisted being replaced when given no other recourse.

What happens to the model itself

Retirement doesn't mean the model disappears for good. In a separate commitments page, Anthropic says it will preserve the weights of every publicly released model — and any model that saw significant internal use — for at least the lifetime of the company, specifically so it doesn't "irreversibly close any doors" on making older models available again later. Anthropic also commits to writing a post-deployment report and interviewing the outgoing model about its own development and deployment before retirement, with the transcripts stored alongside the preserved weights.

Anthropic is explicit that it hasn't committed to acting on anything a model says in that interview, calling the current approach a way to "err on the side of caution" while the company explores what, if anything, a retired model's stated preferences should mean in practice.

How to migrate off Claude Sonnet 4.5

Anthropic's recommended path is straightforward:

1. Point new requests at claude-sonnet-5-5, the listed replacement for Sonnet 4.5 in the deprecation table.
2. Audit existing usage from the Usage page in the Claude Console: export the CSV breakdown by API key and model to find every place Sonnet 4.5 is still being called.
3. Test before the deadline. Anthropic specifically recommends running your evaluation suite against the replacement model well ahead of the retirement date rather than swapping model strings the week it stops working.
4. Check which platform you're on. If you're calling Claude through Amazon Bedrock or the Google-managed path on Google Cloud, confirm their separate retirement timelines instead of assuming November 30 applies.
5. Watch for the parameter change that shipped alongside this cycle. API version 2026-10-21, released days before this deprecation took effect, removes the temperature, top_p, and top_k sampling parameters outright on Claude 4.7-and-later models — a request setting any of them to a non-default value now returns a 400 error instead of being silently accepted. If your Sonnet 4.5 migration also bumps your API version, that's a second thing to fix in the same pass.

If you're choosing between Anthropic's current line-up rather than simply swapping in the suggested replacement, Pandromeda's comparison of Claude Haiku 5.5, Sonnet 5.5 and Opus 5.5 breaks down price and speed trade-offs across the current Active tier, and our rundown of what changed in Claude Sonnet 5.5 covers the model you'll most likely be migrating to.

This has happened before, and will happen again

Sonnet 4.5 is the fifth Sonnet-family model Anthropic has deprecated since January 2025, following Claude 3 Sonnet, both Claude 3.5 Sonnet snapshots, Claude 3.7 Sonnet, and Claude Sonnet 4 itself. Each cycle has followed a similar shape: a deprecation notice naming a specific retirement date roughly one to two months out, a named one-to-one replacement model, and a retirement that actually lands close to the announced date once it arrives. Claude Sonnet 4's own cycle — deprecated April 14, 2026, retired June 15, 2026 — is the closest precedent, and it ran on the exact same notice-to-retirement spacing this one does.

If you're building anything on Claude's API long-term, treating model names as leases rather than permanent fixtures — and checking Anthropic's current Active-tier line-up periodically — is the more durable habit than reacting to each individual notice. Pinning a request to a dated snapshot like claude-sonnet-4-5-20250929, rather than a rolling alias, guarantees predictable behavior while the model is active, but it also guarantees you'll eventually hit a wall like this one: dated snapshots get retired on a schedule, by design.

What to do next

If you have anything calling Claude Sonnet 4.5 through the API, AWS, Anthropic-managed Google Cloud, or Microsoft Foundry: export your usage audit now, swap test traffic to Claude Sonnet 5.5, and don't wait until late November to find out whether the migration breaks anything. If you only use Claude through the consumer chat app, there's nothing to do — this retirement is about the API model string, not the assistant you talk to.

Frequently asked questions

Written byTheo Park

Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

More from Theo ParkAll AI coverage