Model Catalog

The Models page (/models) is the catalog of every model your organization can call right now — across all providers, through a single API and a single key. 78+ models are live today across three providers (Anthropic, Google, and OpenAI), with AWS Bedrock shipping next.

Browsing and filtering

• Search by model name
• Filter by provider, mode (chat, embedding, image, audio, video), or release status (new, GA, deprecated)
• Toggle between a card view and a table view

Each model shows its ID and provider, release status, mode, max tokens, input/output cost, capabilities (vision, function calling, system messages), and current health status. Open a model to see its full specifications in a detail drawer.

Favorites & aliases

Star the models you use most to pin them to the top. Models may also expose aliases — friendly routing names you can pass as the model field in the API.

Comparing models

Use Compare to put models side by side on capabilities and cost. For traffic-based comparisons (real latency, throughput, and reliability from your own usage), see the Benchmark report.

Restrictions

Only the organization owner can restrict which models the org may call. Restricted models still appear in the catalog but are marked unavailable, and calls to them are rejected rather than silently routed elsewhere. Promotional-tier models may be locked until you upgrade.

Requesting a model

Don't see a model you need? Use Request Model to ask for it. To use a model you've already fine-tuned with a provider, pass its model ID in the model field of your API requests where that provider supports it — there is no in-dashboard fine-tune upload.

Next steps

• Analytics & Reports — Benchmark models on your real traffic
• Playground — Try any catalog model interactively
• Chat Completions — Pass a model name to the API

FAQ

Which models can I call today, and does browsing the catalog cost anything?

The catalog lists 78+ models that are live right now across three providers — Anthropic, Google, and OpenAI — all reachable through one API and one key. AWS Bedrock is shipping next but is not available yet. Browsing, searching, and comparing in the dashboard is free; you only pay for the models you actually call.

Do I need a separate key or endpoint for each provider?

No. Every model in the catalog is called through the same endpoint with the same key — you just pass the model's ID (or an alias) as the model field in your Chat Completions request. You never manage provider keys yourself.

What is a model alias and how do I use it?

An alias is a friendly routing name a model may expose that you can pass in the model field instead of the full model ID. It's a convenience for keeping your application code stable; not every model has one, so check the model's detail drawer for its available aliases.

Who can restrict which models my organization is allowed to call?

Only the organization owner can change model restrictions. Admins, members, and viewers see the restriction settings as a read-only badge — they cannot toggle a model on or off.

What happens if I call a model that's been restricted or locked to a higher tier?

Restricted models still appear in the catalog but are marked unavailable for that key or team, and calls to them are rejected rather than silently routed elsewhere. Promotional-tier models stay locked until you upgrade your plan.

How do I compare two models before committing to one?

Open Compare to put models side by side on their capabilities and listed input/output cost. For real-world latency, throughput, and reliability measured on your own traffic, use the Benchmark report instead — the catalog compare is spec-and-price only.

How can I tell whether a model supports vision or function calling, and whether it's healthy?

Each model card shows its mode, max tokens, input/output cost, capabilities (vision, function calling, system messages), and a current health status. Open a model to see its full specifications in the detail drawer before you wire it into production.

Can I pin the models my team uses most?

Yes. Star any model to pin it to the top of the catalog so your frequently used models are easy to find. You can also switch between a card view and a table view, and filter by provider, mode, or release status.

Can I request a model that isn't in the catalog, and who's allowed to?

Yes — use Request Model to submit the model name and your email; any signed-in user can send a request. The team reviews it and follows up at the email you provide, typically within 24 hours.

Can I bring my own fine-tuned model?

Yes — if you've already fine-tuned a model with a supported provider, pass its model ID in the model field of your API requests and it routes like any other model. There is no in-dashboard fine-tuning or dataset upload; the fine-tune happens on the provider side and NemoRouter calls the resulting model for you.