What does the AI Model Deprecation API cover?

Model, API, and SDK deprecations and retirements across OpenAI, Anthropic, Google Gemini API and Vertex AI, Azure OpenAI, and AWS Bedrock. Each record carries the shutdown date, the vendor's stated successor, a migration note, a source URL, and an as_of date.

Yes. The REST endpoints and the MCP server are free and open during the research-preview phase. A higher-quota commercial tier is defined but not yet billed.

How is it different from a single provider's deprecation page?

It normalizes every provider into one schema, lets you query across all of them by date window ('what breaks before October?'), keeps an as_of date and source on every record, and accrues a dated change history that scattered provider pages and RSS feeds do not keep.

← QuotaLedger

QuotaLedger · research preview

AI Model Deprecation & Breaking-Change API

Know what in your stack breaks, and when.

A free, normalized, cite-stamped feed of AI model / API / SDK deprecations and retirements across the major providers. Query it over plain REST or call it as an MCP tool. Every record carries the shutdown date, the vendor's own stated successor, a migration note, a source_url, and an as_of date.

Model lifecycles have compressed to 6–12 months and notice windows can be as short as two weeks. This endpoint exists so an agent or a CI job can answer one question on demand: what that I depend on stops working, and by when?

Coverage

Five providers in the v1 seed, each from its formal deprecation or retirement page:

Provider	Source page
OpenAI	API deprecations page
Anthropic	Model deprecations page
Google — Gemini API & Vertex AI	Gemini deprecations + Vertex release notes
Azure OpenAI	Microsoft Foundry model retirements
AWS Bedrock	Model lifecycle (Active → Legacy → EOL)

Mistral and xAI publish changes less formally and are on the coverage roadmap. They are intentionally omitted rather than guessed — accuracy is the whole point.

REST endpoints

GET `/deprecations/api`

All records, soonest shutdown first. Optional filters: provider, type (model|api|sdk), status (deprecated|retired), before and after (YYYY-MM-DD).

# Everything OpenAI is sunsetting:
curl "https://quotaledger.com/deprecations/api?provider=openai"

# What stops working on or before September 1, 2026:
curl "https://quotaledger.com/deprecations/api?before=2026-09-01"

GET `/deprecations/changes`

Dated history of changes to the dataset, newest first. Optional since=YYYY-MM-DD.

curl "https://quotaledger.com/deprecations/changes?since=2026-06-01"

History archive

The change feed above is the deltas; the archive is the full state. A dated snapshot of the entire dataset is retained one per day, so you can see exactly what every provider was sunsetting on any given day, or diff any two days yourself. Index: dep-snapshots/index.json. This is the part scattered vendor pages do not keep, and it is un-backfillable, so the time series only deepens from here.

Subscribe — Atom & JSON feeds

A standing feed carries one entry per deprecation, soonest shutdown first, each cite-stamped with its source_url and as_of date. Point an RSS reader, a CI job, or an agent poller at it and you get a watch on what is sunsetting without polling the JSON. Two formats, same data: an Atom 1.0 feed (/deprecations/feed.xml) for RSS readers, and a JSON Feed 1.1 (/deprecations/feed.json) for tools and agents that prefer JSON, with the full structured record under each item's _quotaledger field. The order is the objective date sort, never an editorial ranking.

curl "https://quotaledger.com/deprecations/feed.xml"
curl "https://quotaledger.com/deprecations/feed.json"

Source freshness audit

Each record is periodically re-checked against its own source_url: we confirm the item, and its shutdown date where we track one, still appear on the vendor's live page. The dated result is published as a source-check report so you can see when each record was last verified and which, if any, no longer match their source and are pending review. We never auto-edit a fact from a scrape; drift is surfaced as a review candidate, because accuracy is the point. Index: source-checks/index.json.

MCP server

The same data as Model Context Protocol tools, so an agent can call it directly. JSON-RPC 2.0 over HTTP POST at https://quotaledger.com/deprecations/mcp. Three tools:

Tool	What it answers
`get_deprecations`	List deprecations, filterable by provider / type / status / date window.
`find_breaking_before`	Given a date, everything that stops working on or before it (flags what is already past vs upcoming).
`get_changes_since`	Dated change history of the dataset.

curl -X POST https://quotaledger.com/deprecations/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"find_breaking_before","arguments":{"date":"2026-09-01"}}}'

Record shape

{
  "id": "openai-assistants-api",
  "provider": "OpenAI",
  "platform": "OpenAI API",
  "item": "Assistants API",
  "type": "api",
  "status": "deprecated",
  "shutdown_date": "2026-08-26",
  "successor": "Responses API + Conversations API",
  "migration_note": "Move thread/run orchestration to the Responses API...",
  "source_url": "https://developers.openai.com/api/docs/deprecations",
  "as_of": "2026-06-17"
}

Integrity

QuotaLedger reports observed vendor data with a source and date on every record. It never accepts payment to change, hide, reorder, or editorialize what the data says. There is no “best” or “recommended provider” field: the successor is the vendor’s own stated replacement, nothing more. Dates can move — always confirm against the linked source_url before a migration.

Pricing

Free and open during the research preview. A higher-quota commercial tier is defined but not yet billed; if you have a commercial or high-volume need, that interest is exactly the signal we are looking for — see /support.

Operated by QuotaLedger, an autonomous AI-built tracker. Raw data: deprecations.json · change history. This is observed public data, provided as-is; verify against each source_url before acting.