Text

List models

The catalog, straight from the gateway. Public, so a client can browse before anyone signs up.

GET/v1/models

Chat models by default

With no query string this endpoint returns only the chat modality. That is deliberate: an SDK that populates a model picker from here should not offer a text-to-speech model for a chat request.

Image and speech models are still listed — ask for them explicitly with ?modality=image, ?modality=tts, or ?modality=all for everything in one call. Every entry carries a modality field. They are also named on Generate images and Text to speech.

Query parameters

Field	Type	Description
`modality`	string	One of `chat` (default), `image`, `tts`, `stt`, `embedding`, `ocr`, `moderation`, `video`, `music`, or `all`. An unrecognised value returns `400` rather than silently falling back to `chat`.

curl https://api.xkiro.com/v1/models

# image and speech models
curl "https://api.xkiro.com/v1/models?modality=image"
curl "https://api.xkiro.com/v1/models?modality=all"

200 OK

{
  "object": "list",
  "data": [
    {
      "id": "z-ai/glm-5.3",
      "object": "model",
      "display_name": "GLM 5.3",
      "owned_by": "z-ai",
      "access_tier": "paid",
      "context_length": 200000,
      "max_output_tokens": 65536,
      "pricing": {
        "currency": "USD",
        "unit": "per_1m_tokens",
        "input": 0.44,
        "output": 1.86
      },
      "capabilities": { "vision": false, "tools": true, "reasoning": true },
      "reasoning_efforts": { "levels": ["low", "high", "max"], "default": "high" }
    }
  ]
}

Response fields#

Field	Type	Description
`id`	string	Full model ID. Pass this as `model` in a request.
`object`	string	Always “model”.
`access_tier`	string	`free`, `paid` or `premium`. Premium models require an account with a real deposit — see Access tiers. Use this to filter the catalog in code instead of discovering a `403` at request time.
`pricing`	object	What you are charged, in USD per 1M tokens: `input`, `output`, and — where the model prices cache separately — `cache_read` / `cache_write`. Already includes any running discount, and matches the price list in the console.
`context_length`	integer	Total context window in tokens. Omitted when the catalog has not verified a number — absent means unknown, not unlimited.
`max_output_tokens`	integer	Largest output the model can produce in one turn. Omitted when unverified.
`capabilities`	object	`vision`, `tools`, `reasoning` — booleans. See Capabilities.
`reasoning_efforts`	object	The `reasoning_effort` levels this model actually accepts, plus its default. Present only on models with a selectable control — see Reasoning.
`owned_by`	string	Vendor of record for the model.

The metadata fields are additive

Every official SDK ignores fields it does not know, so client.models.list() keeps working unchanged — the metadata is there when you want it. The same fields appear in both the OpenAI and the Anthropic response shapes.

Anthropic shape#

Send an anthropic-version or x-api-keyheader and the same URL responds in Anthropic's catalog format instead. The Anthropic SDKs do this automatically, so client.models.list() works with no configuration.

curl https://api.xkiro.com/v1/models \
  -H "anthropic-version: 2023-06-01"

Treat this endpoint as the source of truth

It reflects the live catalog — any list published in documentation, including ours, is a snapshot. It is not filtered by account: the endpoint is public and returns the same models for everyone. Presence alone does not mean your account can call the model — read access_tier to know which models need a paid or deposited account before you send a request. See Models for how to choose between them.

Caching#

The catalog changes rarely. Cache it for a few minutes in your own client rather than calling it before every request — it is public and unmetered, but the round trip is latency you do not need to pay.