Skip to main content

news_analysis

OpenAI and Anthropic Model Deprecation Policy: How Much Warning You Get Depends on a Model Name and a Route

By AInews Desk
OpenAI and Anthropic Model Deprecation Policy: How Much Warning You Get Depends on a Model Name and a Route
6 min read
1056 words

On October 14, 2026, OpenAI added a new entry to its deprecations page, headed "GPT-4 series model deprecations". Teams that built on those models are now asking how long they have. Your warning period depends on two things in your config: the model name you call, and the platform you call it through. Anthropic gives at least 60 days' notice before it retires a publicly released Claude model.

Three OpenAI tiers, and one word decides yours

OpenAI doesn't use one flat notice period. It uses three. Generally available models get at least 6 months. Specialized variants get at least 3 months, and the examples OpenAI lists are chat variants such as gpt-5.1-chat-latest, Codex variants such as gpt-5.3-codex, and deep research variants such as o3-deep-research. Preview models, identified by preview in the model name, may be retired with much shorter notice, such as 2 weeks.

So that one word in a model ID can shrink your runway from half a year to a single sprint. OpenAI names computer-use-preview and gpt-4o-audio-preview as examples of preview models. It says outright that it doesn't recommend preview models for business-critical production workloads unless you can migrate on short notice.

In practice, OpenAI's past notices have sometimes run longer than the floor. In September 2025, it told developers using gpt-4o-realtime-preview models they would be removed from the API in six months. In June 2025, the gpt-4o-realtime-preview-2024-10-01 snapshot got three months. Those are generous notices, not promises. The written policy still allows two weeks.

There are two more clauses to know. If safety or compliance concerns force an early retirement, OpenAI will give as much notice as reasonably possible. And some developers may be able to provision dedicated capacity for continued access after a model's shutdown date through sales. That could be worth the cost if a migration slips.

Anthropic's flat 60 days and four lifecycle stages

Anthropic uses one number: it gives at least 60 days' notice before model retirement for publicly released models, sent to customers with active deployments. Note the qualifier "publicly released."

Its lifecycle has four stages, and each one means something different for your code:

  • Active: fully supported and recommended for use.
  • Legacy: will no longer receive updates and might be deprecated in the future.
  • Deprecated: still functional but no longer recommended. Anthropic names a replacement and sets a retirement date.
  • Retired: Requests to retired models fail.

The word "legacy" means much the same at OpenAI, which uses it for models and endpoints that no longer receive updates and says you can expect that a legacy model or endpoint will be deprecated at some point in the future. On both platforms, treat "legacy" as the moment to start planning. OpenAI has also shown that access can tighten before shutdown: as of January 10, 2024, only existing users of certain older chat models could keep using them.

Q: Does Anthropic's retirement date apply if I call Claude through Bedrock?

Not necessarily, and many multi-cloud teams miss this. Anthropic says the dates on its deprecations page apply to the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed), and Microsoft Foundry. For Amazon Bedrock and Claude Models on Google Cloud (Google managed), it points readers to those platforms' own model tables.

In practice, a team running the same Claude model on the Anthropic API and on Bedrock has to track two schedules. If your migration plan relies on Anthropic's page alone, check that it covers the route your traffic actually takes.

ItemOpenAIAnthropic
GA / public model noticeAt least 6 monthsAt least 60 days
Specialized variantsAt least 3 months—
Preview modelsAs short as ~2 weeks—
How you're notifiedEmail, docs, blog for larger changesEmail and documentation
Self-check methodshutdown_date via List modelsUsage page → Export CSV
Third-party routes—Bedrock, Google-managed set own dates

How to check before the email arrives

OpenAI gives you a field you can check in code. Call the List models endpoint (GET /v1/models) and check each model's optional shutdown_date field. It returns a YYYY-MM-DD string, or null if no shutdown date has been announced. The same field is on the Retrieve a model endpoint. A nightly job that checks this field against the model IDs in your codebase is cheap to build and will catch most surprises.

Anthropic's deprecations page offers an audit instead. Go to the Usage page in Claude Console, click Export, and review the downloaded CSV to see usage broken down by API key and model. The two methods answer different questions. OpenAI's field tells you when a model shuts down. Anthropic's export tells you where you're still calling it. You need both answers. That's especially true for keys that someone created for a prototype and never cleaned up.

A long data center aisle of server racks where most status lights glow green and one rack at the far end sits dark and powered down.A long data center aisle of server racks where most status lights glow green and one rack at the far end sits dark and powered down.

OpenAI's list doesn't stop at models. The Assistants API was a whole interface, and it went through the same process, with the Responses API and Conversations API named as replacements. Deprecation policies apply to the endpoints you build on, not just the models behind them.

Plan for the floor

Here is a practical rule. Budget for the minimum each vendor guarantees, not the notice they gave last time. For an OpenAI model with preview in its name, assume two weeks. For a specialized variant, assume three months. For Claude on Anthropic-operated platforms, assume 60 days. For Claude on Bedrock or Google-managed Vertex, check that platform's table. Both vendors give the same advice, and it costs little to follow: Anthropic recommends testing with newer models well before the retirement date, and OpenAI says its notice periods exist so customers can evaluate recommended replacement models, test application behavior, and complete migrations. Your actual notice period is set by the model names in your config and the platform you call them through, so check both before you rely on any timeline.

Sources

6 min read
•
1056 words

Tags

OpenAI Anthropic model deprecation policymodel deprecation warning periodsAI model sunset scheduleslegacy model discontinuation timeline

Related Articles