Every AI model we track.

A price is per million tokens, blended 3 in : 1 out, before tax. A row shows the cheapest price anyone charges for the model. A price below the maker's is listed only when the provider says why; where nobody sells it for less, the row shows the maker's own price.

Capability is the model's average across LiveBench's seven categories, out of 100, for the exact version and settings we could match; a model without a matched result shows none. The methodology has the evidence and the rules.

A task orders the list by the categories that task leans on, and models without a score follow, newest first. Automation lists only the models that call tools.

The filters live in the address, so a filtered list is a link you can share.