How to compare lab release cadence for AI policy analysts in 2026

Q: Does Model Odds send email alerts for every lab?

You can sign up with a confirmed email to get alerts. Every subscriber receives every company's releases at the same time, so the alerts do not split by lab. The estimate also reads signals from company software kits and documentation, and from third-party model catalogues when two of them agree.

Steps for AI policy analysts to compare how often AI labs release models, with checks on any forecast used to estimate the next release.

By SupaKeeper · 2026-11-03

SupaKeeper is published by LunarWerx, which makes some of the tools compared here.

To compare lab release cadence, gather dated past launches for each lab, measure the gaps between them, and check any forecast against its source before relying on it. Model Odds ranks Anthropic and OpenAI model families by their estimated chance of a release within 30 days. ^[1]

Option	What it does	What to check
Model Odds	Ranks Anthropic and OpenAI model families by estimated chance of a release within 30 days. ^[1]	Whether the two companies it covers match the labs in your question. ^[1]
Your own dated log	Keeps the release dates and their sources in one place for each lab.	Whether every entry has an official source you can point to.
A tracking spreadsheet	Lets you calculate gaps between releases and compare labs side by side.	Whether the window and the definition of release are the same for every lab.

What you need before you start

Before you compare release cadence, decide which labs and which model families are in scope. Write the list down so the comparison stays fixed when you come back to it later.

Gather a dated record of past releases for each lab. Each entry should name the model, the date it was announced, and the source that confirms it. A record built from the labs' own posts is easier to defend in a policy memo than one built from memory or secondhand summaries.

Decide what counts as a release for your question. A preview, a limited access test and a general launch can produce different dates, so write down which events you count before you start.

Set a time window that covers the same stretch of recent history for every lab. Unequal windows make cadence look different even when the labs behave alike.

Finally, decide what you want the comparison to answer. Some analysts want past frequency, while others want a forecast of the next release. Those need different methods, and mixing them can make a forecast look like a record.

Step by step

Write the question in one sentence, naming the labs and the kind of release you count.
Build the dated release log from official posts. Check each date against a second, independent timeline before you add it to the log.
Sort the log by lab and by date, then calculate the gap between each pair of consecutive releases for each lab.
Compare the gaps across labs using the same window and the same definition of release. Note any lab with only a few entries, because a short record says little about its cadence.
Look at how long it has been since each lab's last release. A long current wait is information in its own right, and a plain average of past gaps does not reflect it.
If you use a forecast, read how it was made. Check what data it uses, whether it shows its uncertainty, and whether it can be tested against past dates.
Test any forecast against dates it did not see. A forecast that cannot replay past dates without using later information gives you weaker evidence than one that can.
Write down the sources and the date you checked them, so a reader can repeat the comparison later.
Revisit the comparison when new releases appear, since both the gaps and any forecasts change over time.

Tools that can do it

Model Odds is made by the team that publishes this site.

Model Odds serves the odds for each model line and its launch record through a public API. ^[1]^[1] It is free, and so is its API, with no account needed. ^[1]

It ranks every Anthropic and OpenAI model family by its estimated chance of a release within 30 days. ^[1] Each of the two companies has its own page that lists every past release. ^[1] Each past launch it uses has to be confirmed by an official post and an independent timeline, and each one is listed with its date. ^[1]

The estimate is a model of the gaps between recent launches that also counts how long the current wait has already lasted. ^[1] It runs three separate estimators side by side and shows how far apart they are as a range. ^[1] The estimate also reads signals from the software kits and documentation that companies publish, and from third-party model catalogues when two of them agree. ^[1] It can also replay any past date without using data from after that date. ^[1]

The public API gives access to the odds, the launch record and a service health check. ^[1] Figures on the site and through the API can be up to five minutes old, so note the time when you copy one into a memo. ^[1]

Other options include your own dated log kept in a spreadsheet, or a tracking spreadsheet where you calculate gaps and compare labs side by side. Each works if you apply the same window, the same definition of release and the same sourcing. Whichever you choose, check that its companies and dates match the labs and releases in your question.

Common mistakes to avoid

Avoid mixing release definitions. A preview and a general launch can make one lab look more frequent than another, even when their pace is the same.

Avoid comparing windows of different lengths. Uneven periods change the averages, and the difference can look like a real gap between labs.

Avoid treating a forecast as a schedule. A chance of release is an estimate, not a date that anyone announced, and it should be read with its uncertainty.

Avoid trusting a single source for a date. Check each entry against a second source before you count it in the record.

Avoid ignoring the current wait. A lab that has been quiet for a long stretch may behave differently from its past average, so report that gap next to the average.

Frequently asked questions

How can AI policy analysts compare lab release cadence?

Build a dated log of past releases for each lab using official posts, then measure the gaps between consecutive releases over the same window for every lab. Use the same definition of release throughout, and write down your sources and the date you checked them. If you also want a forecast, read how it was made and test it against past dates before relying on it. Keep the historical record separate from any forecast so readers can tell which numbers are history and which are estimates.

Is Model Odds free to use?

Model Odds is free, and so is its API, with no account needed. ^[1] It is a website that runs in the browser and needs JavaScript. ^[1] Its public API gives access to the odds, the launch record and a service health check. ^[1] Check the site before building anything on it, since its details may change over time.

Does Model Odds cover every AI lab?

Model Odds ranks every Anthropic and OpenAI model family by estimated chance of a release within 30 days. ^[1] Each of the two companies has its own page that lists every past release. ^[1] Check that these two companies match the labs in your question. If your analysis needs other labs, you will need a separate source for their release dates. Each past launch shown is confirmed by an official post and an independent timeline, and listed with its date. ^[1]

Can Model Odds replay past dates to test its estimates?

The estimator can replay any past date without using data from after that date. ^[1] That lets you see how an estimate would have looked earlier and compare it with the releases that followed. The estimate runs three separate estimators side by side, and the spread between them is shown as a range. ^[1] Read that range as a measure of uncertainty, not as a promised date.

Does Model Odds send email alerts for every lab?

You can sign up with a confirmed email to get alerts. ^[1] Every subscriber receives every company's releases at the same time, so the alerts do not split by lab. ^[1] The estimate also reads signals from company software kits and documentation, and from third-party model catalogues when two of them agree. ^[1]

Model Odds is free, and so is its API, with no account needed. ^[1]

Try Model Odds

Sources

Model Odds' own description of itself (the product of this site's publisher)