Find a model
Featured shows the models Gateway currently recommends, in its configured order. Featured status affects prominence, not availability. Every Featured model also appears in the full registry, and searching or filtering searches the full registry.
Search by model name or API ID, or narrow the list by capability. Filters work together and show the models that match; choose Clear filters to return to the full catalog. The registry includes text and speech offerings, so check the capabilities before choosing a model for your task.
The registry loads its catalog directly from Gateway. Every listed model has open weights and is served with zero data retention (ZDR). Gateway enforces both requirements for API requests as well as catalog listings. Closed-weight models are currently unavailable.
The catalog loads when you open the page. Choose Refresh catalog to check again. If the catalog cannot be loaded, try refreshing; the page clears earlier results so they are not mistaken for current availability.
Compare models
Each model appears once with one API ID, preserving its complete version and variant. Gateway decides how each request is served. The registry does not offer alternative IDs for the same model.
Context is measured in tokens. The displayed context and prices describe the usual way the model is served and can differ for a particular request. A missing context limit means it has not been published. Capabilities are everything the model can offer; a request still needs all of its requirements available at once.
A catalog listing does not guarantee capacity for every request. Capabilities do not establish model quality or licensing terms.
Read model specifications
Model size is usually reported in billions of parameters (B). An advertised size follows the publisher’s label; a checkpoint parameter count measures the published weights and may differ from that label. Size alone does not measure quality or speed.
Gateway supplies reference specifications from published sources. Expand Details and sources to see the architecture, evidence links, and individual check dates. The summary keeps size, weights, and license information together. Open weights means model weights are available to download; consult the linked license for the terms. Unpublished facts are omitted. A missing weights label does not establish whether weights are available.
These reference specifications have their own verification date. The catalog check time covers live availability, context limits, and prices. Capability icons accompany text labels in each model card; the long-context label means the model reports a context limit of at least 100,000 tokens.
Read token prices
Input and output prices are shown in USD per million tokens. Input is what you send to the model; output is what it generates. An unpublished price is unknown. A displayed zero means the catalog explicitly reports zero for that token rate.
Prices labeled from are the lowest input and output rates currently available for that model. The two rates need not come from the same place, and a particular request can cost more. Fees outside the token rates are not shown.
These are list rates for the model. They do not show your quota balance or the final cost of a request. Prices and availability can change after the displayed check time.
Use an API ID
Choose Copy API ID on the model card. Gateway handles routing. Use that exact ID with the appropriate API endpoint for its capabilities. Text generation and speech recognition use different request formats.
The API key guide explains how to create a personal key and send a chat request. Browsing the registry does not require a key.