AI Model Error Decoder

Common error patterns we recognize

Different providers return deprecation errors in slightly different shapes. The decoder handles each one — paste any of these patterns and it will identify the model.

{"error":{"message":"The model `gpt-4.1-nano` has been deprecated...","type":"invalid_request_error","code":"model_not_found"}}

{"type":"error","error":{"type":"not_found_error","message":"model: claude-3-haiku-20240307"}}

{"error":{"code":404,"message":"models/gemini-2.0-flash is not found for API version v1beta"}}

google.api_core.exceptions.NotFound: 404 Publisher Model 'projects/.../publishers/anthropic/models/claude-3-7-sonnet@20250219' not found

Why this happens

AI providers retire older model versions on a rolling schedule. OpenAI typically gives 6-12 months of notice; Anthropic uses 12-month deprecation windows; Google often retires preview models more aggressively. Each provider publishes the schedule in their docs, but watching every changelog is impractical for a team.

The decoder works against a continuously synced dataset, so if a model is officially marked deprecated by its provider, it appears here within a day. If you paste an error and we don't find a match, the error usually isn't a deprecation — common alternatives are a typo in the model id, missing API key access to a private preview model, or a region/billing issue.

Privacy

The decoder runs entirely in your browser. Your error message, including any internal model ids, customer ids, or request payloads it might contain, is never sent to a server. The matching logic and the dataset of 194 deprecations are bundled into the page itself.