Skip to main content
Every request names a model by its id: the publisher’s slug, a slash and the model name, such as anthropic/claude-sonnet-5.5. GET /v1/models lists every id with its price, context window, inputs and accepted parameters. Your code can use it to pick a model and check what it supports. The list needs no key.

List models in code

The OpenAI SDK lists the catalog with client.models.list(). Because the SDK always sends a key, pass a real one: an invalid key fails with 401. Your balance doesn’t change the list: at zero, models without :free are listed too, and a call to one fails with 402 insufficient_quota until you add credits. With a key of an organization or workspace that allows only some models, the list holds only those.

Read a model entry

One entry, shortened:
What each field holds:

Filter by endpoint

To list the models one endpoint serves, send its name in the x-naga-supported-endpoint header. The list keeps a model when any of its operations[].endpoints holds that name. A value no model serves returns an empty list.

Free and paid models

A model id that ends in :free costs nothing. Free models share one rate limit per account, and every other model needs a balance of at least 0.05 credits. Both limits are in Limits.

Aliases and dated ids

A model also answers to its bare name and to every name in aliases, alone or after the publisher’s slug: anthropic/claude-opus-5.5, claude-opus-5.5, claude-opus-5-5 and anthropic/claude-opus-5-5 name the same model. Code that calls the OpenAI or Anthropic SDK with their own model names, such as Claude Code, keeps working. Some models have a dated id and a short alias, for example openai/omni-moderation-2024-09-26 and omni-moderation-latest. A request may use either. The model field of the response always holds the model’s id, so pin the id you see there if you need the exact version. A publisher may also go by another slug elsewhere, such as alibaba for qwen or meta-llama for meta. GET /v1/startups lists those spellings in each publisher’s aliases, and a model id accepts any of them before the slash.

Browse the catalog on naga.ac

naga.ac/models lists every model with its lowest NagaAI price, context window and inputs. A model page’s Pricing section adds the developer’s price, or the lowest price elsewhere, struck through where it is higher. The list, the home page and the Pricing section show how far the NagaAI price is below it as a −N% tag. The tag reads up to −N% when a dearer route can answer. You can filter by input and output type, by supported parameter and by publisher, and sort by price, latency, throughput or popularity. Each model page has a copy button for its id and a Try in Playground link. In the Playground, Compare sends one request to up to four models of the same kind. Each answer shows its time and price. A dot next to a model’s name means its requests are failing right now. A yellow dot means 85–95% of requests succeeded in the last hour, and a red dot means fewer than 85%. Point at the dot to see the exact share. A model with little traffic in the last hour is judged on the last 6 hours. A model with fewer than 10 requests in that time gets no dot. The share counts successes against every failure a provider caused, including a provider’s refusal of the content and a provider’s 4xx that reached you. Errors that NagaAI returns before any provider sees the request, such as validation, your key, your balance or your rate limit, are left out. The uptime chart on the model page counts the same way. Supply often recovers within minutes, so the dot can disappear soon after it shows up. Turn on Hide degraded models in the filters to list only models without a dot. The same dot appears on the model page and in the Playground model picker. Each model page also shows how the model has run lately, across every route NagaAI serves it from. Performance has the daily median (P50) time a request takes to complete, for every kind of model. A chat model adds throughput in tokens per second and time to first token. A request that does not stream gets its first token with the whole answer. It opens on Last 7 days; pick Last 30 days for the month. Uptime shows the share of requests that succeeded for each hour of the last 3 days and every 10 minutes of the last 24 hours. Days are UTC, and the performance numbers refresh hourly. Under the prices, the model page shows the weight precision the model is served in, such as fp8 or bf16. Each value comes from a provider’s own documentation or API. When some provider of the model doesn’t publish its precision, the line ends with or not disclosed. A model whose providers publish none has no precision line. NagaAI picks the provider for each request, so a model listed with several values can answer in any of them, unless the request sets routing.precision.