Model lifecycle tracking
Know before your model shuts down.
Pricing tables tell you what a model costs today. They do not tell you when it stops answering. AI Model Overview tracks the status, deprecation date and shutdown date of 418 models across 20 providers — each backed by the provider's own page — so a retirement is something you plan for, not something you find out about.
What ends next
The nearest announced shutdown dates in the dataset. Every date comes from the provider's own announcement — follow a model through to see it.
- Vercel AI Gateway Retired
Gemini 3 Pro Preview
google/gemini-3-pro-previewShut down on
$2.00 in · $12.00 out / M tokens
- Microsoft Azure AI Foundry Deprecated
DeepSeek-R1
DeepSeek-R1Shuts down on in 2 days
— in · — out / M tokens
- Groq Deprecated
Llama 3.3 70B Versatile
llama-3.3-70b-versatileShuts down on in 5 days
$0.59 in · $0.79 out / M tokens
- Groq Deprecated
Llama 3.1 8B Instant
llama-3.1-8b-instantShuts down on in 5 days
$0.05 in · $0.08 out / M tokens
- Moonshot AI Deprecated
Kimi K2.5
kimi-k2.5Shuts down on in 20 days
$0.60 in · $3.00 out / M tokens
- Moonshot AI Deprecated
Moonshot V1 128K
moonshot-v1-128kShuts down on in 20 days
— in · — out / M tokens
Why this keeps catching teams out
Model shutdowns are announced. They are just announced everywhere except the place you look when you pick a model.
-
The end date is never where the price is
Providers announce shutdowns in changelogs, migration guides, blog posts and doc footnotes — each on their own schedule, in their own format. Pricing tables index none of it.
-
Deprecated is not the same as gone
A model can sit in deprecated limbo for months. It still answers requests, so nothing looks wrong — until the retirement date lands and the same call returns an error.
-
Nobody owns the calendar
The person who chose the model is rarely the person on call when it stops. Without a shared list of end dates, migration windows get discovered, not planned.
-
Prices move quietly too
A tariff change rarely breaks anything, so it goes unnoticed — until it shows up as a line on the invoice you have to explain.
Right now 126 of the 418 models we track are already deprecated or retired — 34 on the way out, 92 gone. If any of them are in your codebase, that is a migration you have not scheduled yet.
One record per model, kept honest
We collect model data from every provider, check it against the primary source and publish it as one dataset — organised around the question the other tools skip: how long does this model have left?
-
Lifecycle status as first-class data
Every model carries a status, an announced deprecation date and a shutdown date — not as a footnote, but as the field the whole dataset is organised around.
-
A shutdown calendar you can plan against
See what ends next, sorted by date, with the days remaining spelled out. 25 models currently have an announced shutdown inside the next 90 days.
-
A source behind every claim
Each model links to the provider page the record came from, with the date it was last checked. Prices taken from an aggregator listing are labelled as such, never passed off as an official tariff.
-
In your assistant, through MCP
Connect the dataset to Claude, your IDE or any MCP client and ask about a model where you are already working — no tab, no copy-paste, no stale screenshot.
-
Prices and context, as context
Input and output cost per million tokens and the context window sit on every model, so a migration decision does not need a second tool.
-
Alerts on the models you actually use in development
Tell us which models your systems run on and get told when one of them is deprecated or scheduled for shutdown — instead of checking a list you have to remember to open.
Not another price comparison
There are plenty of good token-price tables. They answer "what is the cheapest model right now?" — a question that resets every week. We answer the one that costs you a weekend: "what am I running that is about to stop working?"
A price table gives you
- Input and output cost per model
- The cheapest route today
- A snapshot with no memory
We add the part that breaks production
- Status, deprecation date and shutdown date on every model
- A dated history of what changed and when
- A first-party source and verification date per record
- Suggested replacements from the same provider
- The whole dataset in your AI assistant through MCP
How it works
-
We read the provider docs so you do not have to
Model listings, pricing pages, deprecation notices and migration guides are collected per provider, checked, and recorded with the source URL and a verification timestamp.
-
Lifecycle changes become dated events
A new model, a status change, a shutdown announcement or a price move is stored as a change with a date — so the question "when did this change?" has an answer.
-
You get the answer where you work
Browse the overview, jump to a model page, or pull the same data into your AI assistant through MCP. Soon, alerts push it to you instead.
Want the details — which sources, how conflicts are resolved, what we deliberately do not do? Read the methodology →
Ask your assistant instead
The dataset ships as an MCP server, so any MCP-capable client — Claude, your IDE, your own agent — can query model status, prices and end dates directly. The answer arrives with the same source links you would get here.
Endpoint: https://mcp.aimodeloverview.com
"Is the model in this repo still supported?"
Deprecated — shutdown announced. Here is the date, the provider's notice, and the active models you could move to.
Free today. Alerts next.
Everything on this site and in the overview app is free: the full model list, every lifecycle date, the sources, and MCP access. No card, no trial clock.
What we are building next is the push side — alerts for the specific models your systems depend on, so you hear about a shutdown without having to check. We will say what that costs when it ships.
Find out today, not on the day it breaks
Look up any model, see how long it has left, and check what to move to — for 20 providers and 418 models.
Open the model overview