How to find every model ID in your codebase

Model ids live in source as plain strings. Search for the id shapes the providers publish, then check hits against the SunsetDiff scan list. A string that never appears in git can still be set in the host environment.

Last reviewed October 20, 2026. Publisher: SunsetDiff.

Steps

Search for the published id shapes. OpenAI uses model strings such as gpt- and dated snapshots. Claude uses claude-. Gemini uses gemini-. Bedrock uses modelId values such as anthropic.claude-.
Search for the argument name. SDKs put the id in a model argument. Azure and LangChain also use a deployment name.
Check retiring ids. The scan page builds a ripgrep pattern from dataset ids that have a shutdown date and are not active. Nothing is uploaded.

Id shapes from the docs

Search the work tree for the shapes those pages use. A match is a candidate, not proof the call still runs.

OpenAI model pages and the deprecations table use ids such as gpt-4o-2024-05-13 and gpt-5.6-sol. Aliases such as gpt-5.6 are also real request strings.
Claude ids look like claude-sonnet-4-5-20250929 or claude-sonnet-5-5. Bedrock prefixes the older ones as anthropic.claude-...-v1:0.
Gemini ids look like gemini-2.5-pro or gemini-flash-latest.

ripgrep

rg -n -e 'claude-' -e 'gemini-' -e 'gpt-' -e 'anthropic\.claude-' -e 'modelId'

Sources: OpenAI deprecations , Claude model IDs , Gemini API models , Bedrock inference API , checked October 20, 2026.

SDK argument names

The AI SDK puts the id in openai('model-id'). LangChain puts it in init_chat_model("...") or model= on ChatOpenAI, ChatAnthropic, and ChatGoogleGenerativeAI. Bedrock's documented LangChain id is the first argument, for example us.anthropic.claude-sonnet-4-6, with model_provider bedrock_converse.

LiteLLM puts the provider string in model or in litellm_params.model. The proxy's model_name can be a private alias that is not the provider id. Read both keys.

Azure calls a deployment. LangChain's documented parameter is azure_deployment, and its sample reads AZURE_OPENAI_DEPLOYMENT_NAME. Search for that name. The OpenAI model id on the Azure schedule is the version behind the deployment, which you will not see if the code only stores the deployment name.

Argument names

rg -n -e 'azure_deployment' -e 'AZURE_OPENAI_DEPLOYMENT' -e 'litellm_params' -e 'init_chat_model'

Sources: AI SDK OpenAI , LangChain models , LiteLLM proxy config , Working with Azure OpenAI models , checked October 20, 2026.

SunsetDiff scan

The scan page lists retiring ids from this dataset: a published shutdown date and a status other than active. Save them as retiring-ids.txt and run the fixed-string search. Word boundaries in the one-line pattern keep a short id from matching inside a longer one.

A hit means that id is in the dataset with a shutdown date. Open the model page for the source. A miss is not a promise the model is still available. Ids in a dashboard, a secret store, or a vendor console are outside the repo. Anthropic's usage export is the check for ids your Claude API key has actually called.

Retiring ids

rg -n -F -f retiring-ids.txt

Source: Anthropic model deprecations , checked October 20, 2026.

Related SunsetDiff pages

All guides

Questions

Does the scan upload the repository?

No. You copy the pattern and run it locally. SunsetDiff does not receive the files.

Why did search miss an Azure model?

The code may store a deployment name. LangChain calls that argument azure_deployment. The model version is on the Azure deployment, in the retirement schedule.