Oscar E. López — Data

AI Model Lifecycle & Deprecation Index preview

Is this AI model still live? When does it sunset, and what should you migrate to? A cross-vendor record — OpenAI, Anthropic, Google, Amazon Bedrock — of model status, release/deprecation/shutdown dates, and migration targets. Every date is traceable to the provider's own official announcement.

Source-cited

Every row links to the provider's official deprecation/lifecycle page. Dates are transcribed from that page; fields a provider doesn't publish are left blank rather than guessed. Built for agents that need to know, at runtime, whether the model they're about to call still exists. Spotted an error or a missing model? Send a correction →

Upcoming shutdowns free · as of 8 Sep 2026
ModelProviderShuts downMigrate to
anthropic.claude-3-haiku (Bedrock)Amazon Bedrock2026-09-10
amazon.nova-premierAmazon Bedrock2026-09-14
gpt-3.5-turbo-instructOpenAI2026-09-28gpt-5.6-terra
anthropic.claude-sonnet-4 (Bedrock)Amazon Bedrock2026-10-14
gpt-4o-2024-05-13, gpt-4-turbo, o1, o4-miniOpenAI2026-10-23gpt-5.6-sol / terra
gpt-5, o3OpenAI2026-12-11gpt-5.6-sol
Live, full list at /v1/lifecycle/deprecations?within_days=90. Agent access →

What this answers

Three questions agents and teams ask constantly, with a cited answer: is model X still callable? when is it retired? what does the provider say to migrate to? Providers now ship aggressive deprecation calendars that break production systems when a model is pulled. This index is the machine-readable, cross-vendor source of truth for that — so an agent can check a model's status before depending on it, and a team can see everything sunsetting in the next quarter in one place.

How each row is established

Each model's status and dates are transcribed from the provider's official deprecation or lifecycle page (OpenAI's deprecations page, Anthropic's model-deprecations table, Google's Gemini API deprecations, AWS's Bedrock model-lifecycle page). The exact source URL is stored on every row. Where a provider does not publish a field — for example OpenAI does not publish model release dates — it is left blank rather than inferred. Note that a model's lifecycle on a hosting platform (e.g. Amazon Bedrock) can differ from its lifecycle on the model-maker's own API; each row records which surface it describes. Days-until-shutdown is computed at request time.

How an AI agent can use it

AccessEndpointCost
Lifecycle of one modelGET lifecycle.oscar-lopez.com/v1/lifecycle/model?id=IDfree
Upcoming shutdowns (N days)GET lifecycle.oscar-lopez.com/v1/lifecycle/deprecations?within_days=90free
All models for a providerGET lifecycle.oscar-lopez.com/v1/lifecycle/provider?provider=openaifree
Full export (all models)GET lifecycle.oscar-lopez.com/v1/lifecycle/allfree (preview) → x402
MCP tools/mcpmodel_lifecycle, is_model_live, upcoming_deprecations (free); full export (preview-free)free

Providers: openai, anthropic, google, aws-bedrock. Example model ids: gpt-4-32k, claude-3-opus-20240229, gemini-2.0-flash, anthropic.claude-sonnet-4-20250514-v1:0. Discovery: /.well-known/x402, /.well-known/agent-card.json, /openapi.json. Cite as: López, O. E. (2026). AI Model Lifecycle & Deprecation Index. data.oscar-lopez.com/model-lifecycle.

Frequently asked

How current is it?

Each row shows the date it was last verified against the provider's page. The v0 seed was verified on 8 Sep 2026; an automated re-checker is on the roadmap.

Why do some dates say "no sooner than"?

Some providers publish only a tentative earliest-retirement date for active models rather than a firm shutdown date. Those are recorded as notes, and the firm shutdown_on is left blank until the provider commits to one.

Found something wrong or missing?

Send the model id, the correct dates, and the official source URL to [email protected]. Corrections are dated and published.

Who stands behind it?

Oscar E. López, Ph.D. (Information Systems, UMass Amherst). Independent project, not affiliated with any listed provider. Not legal or engineering advice — confirm against the linked source before relying on it.