monitoring · living page
Model releases, as observed
Every foundation model the CLIs we track can run, with three dates side by side: the day the vendor says it shipped, the day the vendor's docs first listed it, and the day our benchmark first saw it answer through a real release of Claude Code or Codex. Read live from the same feeds the pipelines publish; nothing on this page is typed in.
Loading the registry… If this line does not change, the registry did not load; the feeds listed below are the source.
Right now
Foundation models
Newest first. Vendor date is the release date the vendor publishes; where a vendor publishes none, it is the day the vendor's models page first listed the model, marked as such. First seen answering is the earliest benchmark run in which a released CLI version answered with that model — it lags the vendor by however long the CLI took to ship support, which is part of what this page is for.
| model | vendor | vendor date | age | first seen answering | state |
|---|---|---|---|---|---|
| Loading /api/models.json… | |||||
CLI releases
The newest version of each CLI the verification gate has admitted to the
registry, and how long ago it shipped. Both CLIs release far more often
than the models do; the model above and the release here are what a
fresh ovm install gives you today.
| cli | latest admitted | released | age |
|---|---|---|---|
| Loading /api/registry.json… | |||
How the dates are observed
Vendor date. A daily job (06:00 UTC) reads Anthropic's
model lifecycle page for model IDs it has not seen and resolves each new
one's release date from the models overview or the API release notes.
The same job reads OpenAI's models page for frontier gpt
slugs; OpenAI publishes no release date there, so a new model is recorded
with the day it was first seen and a note to fill the date in by hand. A
page that stops parsing fails the job rather than reading as a quiet day.
First seen answering. Every CLI version the gate admits is run on a Mac mini against a real account, and each run records which model answered. The earliest such row per model is what the table shows, with the CLI version it happened on.
Feeds this page reads. Check any number against them.
/api/models.json— the model registry;/models.atomis the same data as a WebSub-pushed Atom feed./api/registry.json— the newest admitted version per CLI;/api/claude.jsonand/api/codex.jsoncarry the dates./model-matrix.json— the benchmark's CLI-version × model rows; the benchmark page plots them.