AI Model Watch
About

The instrument panel of record for AI models.

AI Model Watch tracks 292 models across 12 providers — prices, context windows, modality, lifecycle status and deprecation dates — in one place, refreshed daily. It exists because the single most avoidable production incident in AI is a pinned model that silently gets retired.

Methodology

Data is compiled from each provider's official pricing and model documentation. Flagship models are adversarially re-checked against the source before publishing, and every model page links to where its figures came from. Prices are normalized to USD per million tokens so models are directly comparable.

Last full refresh: 2026-09-26.

FAQ

Where does the data come from?

Every model entry is compiled from the provider's own pricing and documentation pages, and flagship pricing is cross-checked against the official source before publishing. Each model page links to the source it was drawn from.

How often is it updated?

A scheduled routine refreshes the catalog and changelog daily, and re-checks prices and lifecycle status. The "updated" date in the footer reflects the last run.

What does "blended price" mean?

A single-number cost proxy: 0.75 × input + 0.25 × output price per million tokens. Real workloads vary, so use the calculator for your own input/output mix.

Is it really free?

Yes. The data and alerts are free. There are no accounts and no paywall.

How accurate is it?

We aim for accuracy and cite sources, but prices and model availability change fast. Always verify against the official provider page before relying on a number for billing or production.