What does the scanner check besides model IDs?

It also flags deprecated request parameters that error or no longer apply on newer models: the thinking budget_tokens field (replaced by adaptive thinking on Opus 4.7/4.8 and Fable 5), sampling parameters that 400 on those models, the deprecated output_format parameter, older tool-version strings, beta headers that are now GA, and Opus 4.7 Fast mode, which is scheduled for removal on 2026-07-24.

Effloow / Tools / Claude Model Migration Scanner

Tool Forge Free

Claude Model Migration Scanner

Paste your code, config, or .env to find retired and deprecated Claude / Anthropic model IDs and API parameters that break after model retirement. Get a dated migration checklist. Everything runs in your browser — nothing is uploaded.

Paste code, config or .env

0 lines

Findings

Findings will appear here.
Paste code and press Scan, or try a sample above.

—

Retired

—

Deprecated

—

Review

—

Total findings

What this scanner checks

Retired model IDs

Model strings that now return an API error, such as claude-3-5-sonnet-20241022, claude-sonnet-4-20250514, and the Claude 2.x family. Each finding names the drop-in replacement.

Deprecated, retiring soon

Models still live but with a scheduled retirement — like claude-opus-4-1-20250805 (retires 2026-08-05) — and Opus 4.7 Fast mode, scheduled for removal 2026-07-24.

Deprecated API params

Request shapes that break or no longer apply on Opus 4.7 / 4.8 and Fable 5: budget_tokens, sampling params, output_format, old tool versions, and GA beta headers.

Retirement data reflects Anthropic's published model-deprecation schedule. Always confirm against the official deprecation docs before a production cutover.

A scan tells you what is broken today. It cannot tell you what breaks next quarter, because a retirement date is not something your monitoring emits — the argument we make in your prompt tooling has a deadline your monitoring can't see. The same trap catches teams on OpenAI, where an unpinned alias quietly moves under you: we traced that in an audit of the gpt-5 alias and the snapshot behind it.

Frequently Asked Questions

Which Claude models have been retired?

Per Anthropic's published schedule, the Claude 2.x and 3.x families are retired, along with Claude Sonnet 4 and Opus 4 (the -20250514 snapshots, retired 2026-06-15) and Claude Haiku 3. A request to any retired model returns an error rather than a completion. This scanner flags every retired ID it finds and names the recommended replacement.

Is my code or API key sent anywhere?

No. The scan runs entirely in your browser with JavaScript. Nothing you paste — code, config, or an .env file — is uploaded to any server. You can disconnect from the network and it still works once the page has loaded. The scanner only reads text to match model IDs and parameter names; it never sends a request to the Anthropic API.

What does it check besides model IDs?

It also flags request parameters that error or no longer apply on newer models: the thinking budget_tokens field (replaced by adaptive thinking on Opus 4.7 / 4.8 and Fable 5), sampling parameters (temperature, top_p, top_k) that 400 on those models, the deprecated output_format parameter, older tool-version strings, beta headers that are now GA, and Opus 4.7 Fast mode.

Can I trust the scanner enough to skip the docs?

Use it as a fast first pass, not a final authority. It catches the common breakage — retired IDs, date-suffixed aliases, deprecated params — that a grep alone misses, but retirement dates change and Anthropic can add new deprecations. Always confirm a production migration against the official model deprecation docs and test one request against your target model.

Claude Model Migration Scanner

What this scanner checks

Frequently Asked Questions

Related reading

Your Prompt Tooling Has a Deadline Your Monitoring Can't See

The gpt-5 Alias Points at a Model OpenAI Deletes Dec 11

OpenAI Models on Bedrock: A Responses API Readiness Check

Find these tools useful?