Your provider bill is one number a month. It will not tell you which app, which environment, or which model spent it. One line of code meters every call across OpenAI, Anthropic, Azure, AWS Bedrock, and Google, then shows you real spend per app, model, and environment — plus cheaper on-par models, and a warning before one you depend on retires.
Free under $300/mo of tracked spend, no card. The demo needs no account, and neither does the catalog, the API, or the MCP server.
Sources are re-checked at least every 24 hours. Status API
Search and filter the full feed by provider and lifecycle state, with deprecation dates, retirement dates, and recommended replacements.
Coding agents learned their model ids from a training snapshot, so they confidently write ids that stopped answering months ago. The LLMIntel MCP server adds a live lookup before the code lands: is this id still valid, when does it stop working, and what replaces it.
{
"mcpServers": {
"llmintel": {
"command": "npx",
"args": ["-y", "@llmintel/mcp"]
}
}
}DO NOT USE — this model is retired;
API calls to it fail.
Provider: anthropic
Lifecycle state: retired — retired; calls fail
Deprecated: 2026-04-14 (105 days ago)
Retirement: 2026-06-15 (43 days ago)
Source: docs.anthropic.com/en/docs/about-claude
/model-deprecations
Provider's own term: "Retired"Anyone can scrape a table. The hard part is normalizing six vocabularies into one state machine and being right often enough that you'd page someone on it.
OpenAI says “shutdown”, Anthropic runs a four-state lifecycle, Azure publishes a retirement table, Bedrock says “Legacy”. All of it maps to one documented state machine: active → deprecated → retiring → retired.
Every record keeps the provider's verbatim term and a link to the exact page it came from. Parser changes go through a human review queue before publishing, so a page restructure never silently drops a retirement.
Every collector run is recorded whether or not anything changed, so “last checked” is always honest. See it live at /v1/status.
The catalog tells you what's changing. Add one line of telemetry and it tells you what it's costing. instrument(new OpenAI()) meters the models your app actually calls (metadata only, never prompts) and turns that into a live cost breakdown by app, environment, and tag. Switch suggestions are priced against your own token mix, so you get “save ~$1,840/mo” rather than “80% cheaper”.
Free under $300/mo of tracked spend, and retirement alerts for the models you actually run are free on every plan.