Skip to content

Swap retiring AI models before they break production.

ModelExpiry finds the model IDs in your repos, checks each one against the providers’ retirement schedules, and opens a pull request to the official replacement.

$ npx modelexpiry scan
Install the GitHub App
Example: the CLI scanning a sample repository.
➜ npx modelexpiry scan
412 files · 1.8 s · 6 model IDs
deprecatedgpt-4o-2024-08-0624d
deprecatedclaude-sonnet-4-2025051462d
legacygemini-2.0-flash-00183d
retiredgpt-4.5-previewgone
Run npx modelexpiry fix

npx modelexpiry fix

The ID changes. Nothing else does.

Locally, the CLI previews each swap and applies it in place. On GitHub, the App opens the same change as a pull request, with a checklist for your evals.

Example: the pull request the GitHub App opens for a retiring model.

Replace gpt-4o-2024-08-06 with gpt-5.1

Openneeds-eval
− model: "gpt-4o-2024-08-06",
+ model: "gpt-5.1",
Model lifecyclepassed

GitHub App · GitHub Action

A check on every pull request, one report per repo.

The check fails only when a pull request adds a model that is on its way out. Existing debt goes to a single Issue, kept up to date.

Example: the Model lifecycle check failing on a pull request that adds a retiring model, and the retirement report Issue of a sample repository.
Model lifecycleFailing after 3s · 1 retiring model added

This pull request adds gpt-4o-2024-08-06, which OpenAI shuts down on 2026-12-08.

src/summarize.py
18+ MODEL = "gpt-4o-2024-08-06"
Deprecated, shuts down in 24 days. Use gpt-5.1.
Open fix PRView details

Model retirement report #57

Open Updated by modelexpiry 2 h ago

ModelShutdownFix
gpt-4o-2024-08-06in 24 days
claude-sonnet-4-20250514in 62 days
gemini-2.0-flash-001in 83 days
gpt-4.5-previewcalls fail

Start free. Upgrade when you outgrow 3 repos.

Billing period

Questions

Do you store my code or send it to an LLM?

No. The App reads each repo in memory and discards it; we keep the path, the line and the model ID. The CLI and the Action send nothing at all.

Will you merge the pull request for me?

Never. ModelExpiry opens the pull request, your own CI runs on it, and you decide. Close it to snooze the alert until seven days before the shutdown.

Where do the retirement dates come from?

From each provider’s official deprecation pages. Every date keeps its source link and the time we fetched it. New announcements go out within hours; a change that contradicts a published date waits for a person to check it.

What if a model I use is not in the registry?

Then it is never reported: an ID we do not know is not a finding. New models are added as their providers publish them.