Claude Sonnet 4.5 Retirement: Key Dates and How to Migrate

Anthropic has set a firm end date for Claude Sonnet 4.5 on its API. The model was marked deprecated on September 30, 2026 and is scheduled for hard retirement on November 30, 2026, after which every request sent to it will fail outright. Anthropic's recommended replacement is Claude Sonnet 5.5 (API model ID claude-sonnet-5-5), the newer, faster Sonnet-tier model the company shipped at the end of September 2026.

What happened to Claude Sonnet 4.5

Claude Sonnet 4.5 is one of Anthropic's mid-generation Sonnet models, originally released under the snapshot ID claude-sonnet-4-5-20250929. According to Anthropic's official model deprecations page, the company notified developers on September 30, 2026 that the model's status had changed from "active" to "deprecated," with a tentative retirement date of November 30, 2026. In Anthropic's own lifecycle terminology, a deprecated model is "still functional but no longer recommended," while a retired model is "no longer available for use" — requests to it simply fail.

Anthropic uses a four-stage lifecycle for every Claude model: active, legacy, deprecated, and retired. "Legacy" models stop receiving updates but keep running with no fixed end date; "deprecated" models get an explicit replacement recommendation and a retirement date on the calendar; "retired" models stop accepting traffic entirely. Claude Sonnet 4.5 has now moved into that deprecated stage, with the clock running toward a confirmed retirement date rather than an open-ended one.

The announcement follows a pattern Anthropic has repeated through 2025 and 2026: a new flagship or mid-tier model ships, and a couple of generations later the oldest Sonnet or Opus snapshot in active use gets a deprecation notice and a 60-day countdown. Claude Sonnet 4.5 was superseded first by Claude Sonnet 5 and then, on September 28, 2026, by Claude Sonnet 5.5 — the model Anthropic is now steering all Sonnet 4.5 traffic toward.

Key dates for Claude Sonnet 4.5's retirement

The two dates that matter for anyone with code pointed at claude-sonnet-4-5-20250929 are the deprecation notice and the retirement cutoff. Anthropic's documented policy is to give at least 60 days' notice before retiring any publicly released model — the Sonnet 4.5 timeline runs 61 days, in line with that commitment.

Those dates apply specifically to Anthropic-operated infrastructure: the Claude API, Claude Platform on AWS, Claude Models on Google Cloud when it's Anthropic-managed, and Microsoft Foundry. Amazon Bedrock and the Google-managed version of Vertex AI are partner-operated and set their own retirement schedules, so a team running Claude Sonnet 4.5 through Bedrock should check Bedrock's own model table rather than assume the November 30 date carries over automatically.

Why Anthropic is retiring Claude Sonnet 4.5

Anthropic's documentation is candid about the trade-off. The company states it currently deprecates and retires older models "to ensure capacity for new model releases" — in effect, running fewer model versions in production frees up serving capacity for the newest ones. Anthropic acknowledges the downsides of this approach directly: developers who have tuned workflows around a specific model have to migrate, researchers running comparative studies lose access to a fixed baseline, and retiring a model early raises what Anthropic itself describes as "safety- and model welfare-related risks."

To soften those costs, Anthropic has published a standing set of commitments on model deprecation and preservation, under which it says it intends to preserve retired models' weights for the long term and hopes to eventually make older models publicly available again, rather than deleting them outright when they're pulled from the live API.

What breaks on November 30, 2026

The mechanics are simple: before the retirement date, calls to claude-sonnet-4-5-20250929 keep working as before, just flagged as using a deprecated model. After November 30, 2026, any request naming that model ID will fail. For a production application, that typically shows up as:

• API error responses (model-not-found or equivalent) wherever the model ID is hardcoded in backend code, serverless functions, or CI/test configs
• Broken behavior in third-party tools, internal chatbots, or agent frameworks that pin a specific Claude snapshot rather than an alias
• Silent failures in scheduled jobs or batch pipelines that don't surface errors visibly until someone checks logs
• Inconsistent results if only some environments (staging vs. production, or one region vs. another) were updated before the deadline

None of this is unique to Sonnet 4.5 — Anthropic's full deprecation history shows the same failure mode for every retired snapshot, from the Claude 1 and Instant models retired in 2024 through Claude 3.7 Sonnet, Claude 3 Haiku, and Claude Sonnet 4 and Opus 4, which were all retired earlier in 2026. The difference this time is simply which model ID developers need to check for.

Claude Sonnet 5.5 vs. Claude Sonnet 4.5: what's different

Anthropic's recommended migration target, Claude Sonnet 5.5, is two full generations newer than Sonnet 4.5. Per Anthropic's launch post, Sonnet 5.5 is pitched as a faster, cheaper companion to the flagship Claude Opus 5.5, aimed at well-scoped everyday coding, bug-fixing, and document- or spreadsheet-generation tasks. Anthropic says it is more than 30% faster at generating output than the prior Sonnet 5 model and typically needs fewer tokens to complete the same task, translating into savings of up to 30% per task even though its headline per-token input and output prices are unchanged from Sonnet 5, as listed on Anthropic's pricing page. Cached prompt reads on Sonnet 5.5 are priced at $0.10 per million tokens — half of Sonnet 5's cache-read rate. Full details, including early-customer feedback from companies such as Epic Games, are in Anthropic's Claude Sonnet 5.5 announcement.

Two capability changes are worth flagging for teams migrating directly from Sonnet 4.5. First, Sonnet 5.5 is the first Sonnet-tier model to ship with cybersecurity safeguards similar to those on Anthropic's most capable models, plus classifiers meant to prevent reasoning-extraction attacks; some higher-risk cybersecurity-flavored requests may now route to a different model rather than Sonnet 5.5 itself, though Anthropic says routine bug-fixing is unaffected. Second, on the API side, Sonnet 5.5 drops support for disabling extended thinking outright and for forcing tool choice, among other changes documented in Anthropic's dedicated Sonnet 5.5 migration guide, which explicitly covers upgrade paths from Sonnet 4.5, Sonnet 4, Sonnet 5, and Claude 3.7 Sonnet. Sonnet 5.5 is available with zero data retention on AWS, Google Cloud, and Microsoft Azure, which may matter for regulated workloads moving off Sonnet 4.5.

How to migrate off Claude Sonnet 4.5

Anthropic's own guidance, spread across its model-deprecations and migration-guide pages, boils down to a short checklist:

1. Find every place the old model ID is used. In the Claude Console's Usage page, exporting a CSV breaks API usage down by key and model, which is the fastest way to find stray calls to claude-sonnet-4-5-20250929 across different services and environments.
2. Swap the model ID to claude-sonnet-5-5. For teams using Claude's managed-agents platform, Anthropic says no changes beyond updating the model name are required.
3. Read the model-specific migration guide before shipping. Because Sonnet 5.5 changes how extended thinking and forced tool choice work, a plain model-ID swap can produce 400 errors on requests that relied on now-removed parameters. The official Sonnet 5.5 migration guide lists these breaking changes model by model.
4. Re-test before November 30. Anthropic recommends thorough testing against the replacement model well ahead of a retirement date, rather than waiting for production traffic to start failing.
5. Watch Bedrock and Vertex separately. If any workloads run on Amazon Bedrock or the Google-managed flavor of Vertex AI, confirm those platforms' own retirement timelines rather than assuming Anthropic's November 30 cutoff applies there too.

For teams already deep into agentic workflows, this kind of model churn is part of a broader shift in how AI coding agents get maintained in production — pinning a model ID without a deprecation-monitoring process is an increasingly common source of unplanned downtime.

Part of a recurring pattern

Sonnet 4.5's retirement isn't an isolated event. Anthropic's deprecation history page shows a steady cadence of similar notices through 2025 and 2026: Claude Sonnet 4 and Claude Opus 4 were both deprecated on April 14, 2026 and retired June 15, 2026; Claude Opus 4.1 was deprecated June 5, 2026 and retired August 5, 2026; and Claude 3.7 Sonnet was deprecated in October 2025 and retired in February 2026. In nearly every case, the "recommended replacement" column points to whichever Sonnet or Opus model is newest at the time — Claude Sonnet 5.5 is currently the listed replacement for Sonnet 4.5, Sonnet 4, and even the much older Claude 3.7 Sonnet and Claude 3.5 Sonnet models.

That cadence has implications beyond Sonnet 4.5 specifically. Anthropic's broader business has also been under scrutiny this year — the company's IPO valuation was cut after a stock sell-off — and a faster model-refresh cycle is one visible side effect of competitive pressure in the market, alongside new agent products like Claude Cowork. Developers building on Claude should expect this rhythm to continue: a new model ships, an older one gets a deprecation notice roughly 60 days before it goes dark, and the model-status table gets a little shorter.

Who actually needs to act

Not every Claude user needs to do anything. This retirement only affects developers and businesses calling the Anthropic API (or AWS/Google Cloud/Foundry equivalents) directly with the specific model ID claude-sonnet-4-5-20250929 hardcoded somewhere in their stack. Consumer-facing products like claude.ai are unaffected, since Anthropic controls which model version serves end users there and routinely upgrades it behind the scenes. The people who need to check their code are:

• Backend engineers maintaining apps, chatbots, or internal tools that call the Messages API directly
• Teams running batch or scheduled jobs against the API
• Anyone maintaining SDKs, middleware, or proxy layers that default to a specific Claude snapshot
• Vendors who built products on top of Claude Sonnet 4.5 and haven't yet validated against Sonnet 5.5

What happens next

Between now and November 30, 2026, Claude Sonnet 4.5 will keep working, but Anthropic's own warning is explicit: deprecated models are "likely to be less reliable than active models." Developers have roughly three weeks from this notice's publication date to validate Sonnet 5.5 as a drop-in (or near-drop-in) replacement, update any hardcoded model IDs, and re-run test suites against the new model's thinking and tool-use behavior. Anthropic says it will continue notifying customers with active deployments ahead of future retirements by email and through its documentation, so the model-status table on its deprecations page remains the fastest way to check whether any other model a team relies on is next in line.

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