Model Catalog
Every AI model we track — what it costs, how much context it holds, when it shipped — merged nightly from models.dev and LiteLLM, plus intelligence scores from Artificial Analysis. Below the catalog sits the roster: the models an org has approved for each kind of work.
Loading the latest publish run…
Published as static JSON:
catalog.json,
receipt.json,
rosters/geldentech.json.
CORS is open — read them from anything.
Roster — who we've approved for what
The catalog is facts; the roster is policy. It is edited by decision, never by the nightly job, and it names a ranked fallback order per role rather than a single model — so a role survives a model being retired. Skills ask the roster “who does X for us today?” instead of hardcoding a model name.
Loading the roster…
Catalog
Prices are US dollars per million tokens. Click a column heading to sort; untracked values always sort to the bottom, because an absence is not a small number.
A dash (—) means no source publishes that value for that model — hover it for the reason. It never means zero, and it never sorts as zero. Whether a whole column is available at all depends on its source: the status chips at the top of this page say which feeds answered on the last run.
Loading the catalog…
How it is built
A nightly job on the platform fetches three public feeds and merges them by canonical
model id (creator/model-slug). On conflict: pricing follows LiteLLM (fastest
to update on launch day), specifications follow models.dev, and intelligence and speed
come only from Artificial Analysis. Every merged field records where it came from.
Source rows that cannot be mapped to a canonical id are not dropped silently — they
land in unmapped[] in the receipt, and the count is shown at the top of this
page. First-party API models only: Bedrock, Azure Foundry and Vertex variants are
deliberately out of scope.