Last verified

Models AI glossary

What is model deprecation?

Model deprecation is a vendor's announcement that a model is no longer recommended and will be retired on a set date, after which requests to it fail; the vendor usually names a replacement.

At a glance

  • How it works: The model keeps working after deprecation, with a retirement date attached; on that date its API identifier stops answering and calls return errors.
  • Why it matters: Every application pinned to a model inherits a deadline — and the replacement can price, tokenize and behave differently.
  • What it costs: Migration work, plus a new rate card: the successor may be cheaper or dearer per token, and may count your prompts as more tokens.
  • Related to: LLM, open-weight model, tokenizer, price per million tokens.

The model lifecycle

Anthropic's documentation spells out the stages most vendors follow. Active: fully supported and recommended. Legacy: no longer updated and may be deprecated in future. Deprecated: still functional but no longer recommended, with a recommended replacement and an assigned retirement date. Retired: no longer available — requests to it fail.

Anthropic commits to at least 60 days' notice before retiring a publicly released model and notifies customers with active deployments by email. Its recent schedule shows the pattern: Claude Sonnet 4.5 was deprecated on September 30, 2026 with retirement set for November 30, 2026 and Claude Sonnet 5.5 named as the replacement. Partner platforms set their own dates — Anthropic notes that Amazon Bedrock and Google Cloud can retire the same model on a different schedule.

What a retirement costs you

The direct cost is engineering time: changing the model identifier, re-running evaluations, adjusting prompts that relied on the old model's behaviour. Anthropic recommends testing the replacement well before the retirement date and warns that deprecated models are likely to be less reliable than active ones in the meantime.

The indirect cost sits in the rate card. A replacement often has a different per-token price, and it can have a different tokenizer: Anthropic documents that its models from Opus 4.7 onward count the same text as roughly 30 percent more tokens than earlier ones. A migration at an unchanged list price can still raise the bill. Request parameters can change too — Anthropic's newest models reject non-default temperature, top_p and top_k values.

How our index handles retired models

Our price index keeps retired and deprecated models rather than deleting them, each with its status and the date it was set, and records the vendor's named replacement as a link from the old model to the new one. Pages that point at a retired model follow that link to the current successor automatically, so a retirement is a one-row change in our data. The panel below counts the lifecycle states across the whole index.

In our indexModel lifecycle in our index

Current 180 of 344 rows
Deprecated 44 still callable, end date set
Legacy 22 off the price table
Retired 92 134 archived rows name a successor

Counted at build over all 344 rows of our price index, archive included. A status comes from the vendor's own deprecation or models page. How we collect prices.

Staying ahead of deprecations

  • Keep the model name in configuration, not scattered through code, so a migration is one change.
  • Audit usage by model — Anthropic's console exports usage broken down by API key and model, which shows where an old model is still called.
  • Maintain an evaluation set so a replacement can be judged in hours, not weeks.
  • Consider open weights where reproducibility matters: a copy you host cannot be retired out from under you.

Frequently asked.

Short answers to the questions that come up next.

Q · 01 What is the difference between deprecated and retired? +
A deprecated model still works but has a retirement date and a recommended replacement. A retired model no longer answers — requests to it fail.
Q · 02 How much notice do vendors give before retiring a model? +
Anthropic commits to at least 60 days for publicly released models and emails customers with active deployments. Other vendors set their own policies; check their deprecation pages.
Q · 03 Will my costs change when I migrate to the replacement? +
Often. The replacement may have a different per-token price and a different tokenizer — Anthropic's newer tokenizer counts the same text as about 30 percent more tokens — so re-price the workload, not just the rate.

Sources · read on the date shown