LLM Model Sunset Scanner

Find OpenAI, Claude, Gemini and Azure OpenAI model IDs in your code or automation workflows that are already retired or scheduled to shut down, and get the provider's recommended replacement for each one.

OpenAI's next large batch of shutdowns is on 23 October 2026. It includes gpt-3.5-turbo, gpt-4, gpt-4-turbo, gpt-4o-2024-05-13, gpt-4.1-nano, gpt-image-1, o1, o1-pro, o3-mini, o4-mini and several fine-tuned bases (OpenAI deprecations). More shutdowns follow every few weeks from OpenAI, Anthropic and Google. On those dates, requests to the retired models start failing.

What it scans

• Public GitHub or GitLab repositories: up to 5 per run. The Actor downloads the repository archive (50 MB limit) and reads source, config and workflow files. node_modules, build output, lock files and binaries are skipped.
• Pasted content: source code, or an exported n8n, Make or Zapier workflow (JSON). For workflow JSON, each hit also names the workflow node or module that holds the model.
• Raw file URLs: for example a workflow JSON on raw.githubusercontent.com or in a gist.
• Model IDs (lookup mode): check a list of IDs directly, without any code. This is also useful for AI agents that want to check a model name before they write it into code.

What you get

One dataset row per hit:

Field	Example
modelId	gpt-4-turbo
provider	OpenAI
file, line, snippet	app/llm.py, 42, model="gpt-4-turbo"
workflowNodes	OpenAI Chat Model (@n8n/n8n-nodes-langchain.lmChatOpenAi)
shutdownDate	2026-10-23
daysLeft	11 (negative means already retired)
severity	retired, critical (≤30 days), soon (≤90), scheduled
replacement	gpt-5.6-sol (as named by the provider)
sourceUrl	link to the official deprecation page
notes	aliases, platform differences, conflicting dates on the page

The OUTPUT record in the key-value store holds a summary: the earliest deadline, a per-model roll-up, and a migration checklist you can paste into an issue.

Where the dates come from

Dates are taken only from the providers' own deprecation pages:

• OpenAI: https://developers.openai.com/api/docs/deprecations
• Anthropic (Claude API): https://platform.claude.com/docs/en/about-claude/model-deprecations
• Google Gemini API: https://ai.google.dev/gemini-api/docs/deprecations
• Google Vertex AI: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/model-versions
• Azure OpenAI: https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule

Each run fetches these pages first. If a page can't be reached or parsed, the Actor uses a bundled copy and says so: the dataSources field in OUTPUT shows, for each provider, whether the data was live or bundled and how old it is. The bundled copy was taken on 12 October 2026.

Only firm dates count. Tentative wording such as "not sooner than" or "no shutdown date announced" is not reported as a deadline.

How matching works, and its limits

• Exact IDs are matched as written in the provider's table. gpt-4 does not match gpt-4o.
• Aliases are resolved where the provider's naming makes them clear. For example, claude-sonnet-4-5 resolves to claude-sonnet-4-5-20250929, and ft:gpt-3.5-turbo-0125:org::id resolves to its base model. These rows are marked matchType: alias or normalized.
• Platform differences are real. The same Gemini ID can have different dates on the Gemini API and on Vertex AI. For example, gemini-2.5-flash retires on Vertex AI on 20 October 2026, but the Gemini API page lists no shutdown date for it. The Actor picks the platform from clues in each file (Vertex SDK imports, Azure endpoints, n8n node types) and lists the other platforms' dates in otherPlatforms.
• Azure OpenAI dates apply only when a file looks like Azure code. Azure retires each model version separately. When the file doesn't show the version, the Actor shows the earliest upcoming date and lists every version.
• Amazon Bedrock and Claude on Vertex AI set their own retirement schedules. For those IDs the Actor shows the Claude API date with a note; check your platform's page before acting.
• Model names built at runtime (for example from an environment variable or a database) can't be seen in the code. Look those up in lookup mode.
• Model IDs in docs (.md, .txt) are skipped unless you turn on Include docs.

Pricing (pay per event)

• $0.20 per scan: one repository, one raw file, or one block of pasted content. You are not charged if the repository can't be downloaded or nothing readable is found.
• $0.005 per model ID in lookup mode.

Apify's standard Actor start event also applies. Set a maximum charge on the run if you want a hard cap; the Actor stops before going over it.

Example input

{
  "repoUrls": ["https://github.com/your-org/your-app"],
  "modelIds": ["gpt-4", "claude-sonnet-4-5", "gemini-2.0-flash"],
  "horizonDays": 180
}

Privacy

The Actor reads only the public repository, file or text you give it. It doesn't read git history or commit authors, and it stores only the hits (file path, line, a snippet of up to 160 characters) in your own run's dataset.

Disclaimer

This tool is not affiliated with OpenAI, Anthropic, Google or Microsoft. Providers can change dates, so check the linked official page before you plan a migration. If a date looks wrong, open an issue on the Actor page and include the model ID.