Build with the API

List the models you can call

GET /v1/models lists the models your key may call. Use it to find a model's id, check what it can do, and read what it costs, from code.

curl https://sven.swedishembedded.com/v1/models \
  -H "Authorization: Bearer $SVEN_API_KEY"

What the list holds#

Clients written for OpenAI read the list unchanged, and each entry carries more:

Field	Meaning
`id`	The name you pass as `model`.
`name`, `description`	What people call the model, and a line about it.
`context_length`	How many tokens of conversation it can take; `null` where unknown.
`architecture`	What it reads and writes: `input_modalities`, `output_modalities` and the tokenizer.
`top_provider.max_completion_tokens`	The longest answer it can write; `null` where unknown.
`supported_parameters`	The request parameters the model supports.
`capabilities`	What you can do with it: `chat`, `tools`, `embeddings`, `images` or `transcription`.
`pricing`	What it is sold for, in exact US dollars: see below.
`zero_data_retention`	`true` when a call to it keeps no data anywhere.

The list also carries default_model at the top: the model your account's sessions start with when none is named.

A model that has no price is absent: it is not for sale. A restricted key sees only the models it may call, and a model served only by providers that may keep data is marked needs_data_sharing until the person the key acts for allows it (see Models and data).

Prices#

pricing states what the model costs, in exact US dollars per unit, as text so that no digit is lost:

prompt is the price per input token (per second of audio, for a speech-to-text model).
completion is the price per output token.
image is what a typical image costs, for image models: about a megapixel, a short prompt and no reference image. A larger image or one with references costs more.
image_max is the most one image is ever charged; a call is held at it before it runs.
Fees that are not charged are "0".

Read prices from here or from the pricing page rather than copying them into your code: they are the most a call is charged (many cost less, see How calls are charged), and they change in one place.

Pick a model by what it does#

Choose by capabilities: a chat model for chat completions, an embeddings model for embeddings, images for images and transcription for speech to text. A model used for something it is not sold for is refused with a 400 or 404.

Signed in, the Models page shows the same list with prices and which models keep no data.