Figwick Your AI. Your rules. Fig Jar free stuff Join the waitlist

White paper

When the model changes under you

How cloud AI providers retire models, what breaks, and how small firms can plan for it.

For small firms and their IT providers.

  • 13 min read
  • 12-page PDF
Get the PDF See what's inside
Cover of the white paper

Abstract

Cloud AI providers retire models on their own schedules, sometimes with as little as two weeks' notice. After the retirement date, requests to the old model either fail or go to a different model. This paper explains how Anthropic, OpenAI, Microsoft Azure and Amazon Bedrock retire models, what that breaks in a small firm, and the costs that don't appear on an invoice. It ends with a planning checklist and the trade-offs of running some models yourself.

Highlights

  • Anthropic gives at least 60 days' notice before a model is retired. After that, requests to it fail.
  • OpenAI gives at least 6 months for generally available models, and as little as 2 weeks for previews.
  • On Azure, a model retires 18 months after launch, and standard deployments switch to a replacement automatically.
  • On Amazon Bedrock, some models get 45 days' notice, and idle accounts can lose access to a retiring model sooner.
  • The checklist covers inventory, owners, pinned versions, testing, notices and time in the budget.

Contents

  1. What model retirement is
  2. How the big providers retire models
  3. What breaks when a model is retired
  4. Costs that don't appear on an invoice
  5. A planning checklist
  6. Running models yourself
  7. About Figwick

Sources

  1. Model deprecations Anthropic, kept current, read September 25, 2026 https://platform.claude.com/docs/en/about-claude/model-deprecations
  2. Amazon Bedrock model lifecycle (models launched before 2026-09-07) Amazon Web Services, kept current, read September 25, 2026 https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle-legacy.html
  3. Deprecations OpenAI, kept current, read September 25, 2026 https://developers.openai.com/api/docs/deprecations
  4. Foundry Models lifecycle and support policy Microsoft, kept current, read September 25, 2026 https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirements
  5. Amazon Bedrock model lifecycle Amazon Web Services, kept current, read September 25, 2026 https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html
  6. Open and closed model capability gap Epoch AI, May 2026 https://epoch.ai/data-insights/open-closed-eci-gap
  7. Exposed Ollama hosts form an anonymous AI network SentinelOne and Censys, January 2026 https://www.sentinelone.com/labs/silent-brothers-ollama-hosts-form-anonymous-ai-network-beyond-platform-guardrails/