API reference

Models

List the models a key can call, with pricing and limits.

The models endpoint returns every model the calling key may use, with its limits, capabilities and prices. The same catalog is browsable in the dashboard.

List models

GEThttps://odysseyapi.tech/v1/models

curl https://odysseyapi.tech/v1/models \  -H "Authorization: Bearer $ODYSSEY_API_KEY"

{  "object": "list",  "data": [    {      "id": "anthropic/claude-sonnet-5",      "name": "Claude Sonnet 5",      "object": "model",      "created": 1767225600,      "owned_by": "anthropic",      "kind": "chat",      "context_window": 200000,      "max_output": 64000,      "capabilities": ["reasoning", "vision", "tools", "json", "long-context"],      "pricing": {        "input_per_m": 3.0,        "output_per_m": 15.0,        "cached_input_per_m": 0.3,        "per_image": null      }    }  ]}

If the key is limited to specific models, only those are returned. That makes this endpoint a quick way to check what a key can do without reading the dashboard.

The model object

Field	Type	Description
`id`	string	The value to send as model, formatted as author/slug.
`name`	string	Display name, for interfaces.
`context_window`	integer	Maximum tokens across the prompt and the response.
`max_output`	integer	Maximum tokens the model will generate in one response.
`owned_by`	string	The author that trained the model.
`kind`	"chat" \| "image"	Chat models run on chat completions, responses and messages; image models on image generation.
`capabilities`	string[]	Any of reasoning, vision, tools, json and long-context.
`pricing`	object	US dollars per million tokens, or per image for image models.

Pricing

One price per model. What a request costs is covered in usage and billing.

Cached input

Where a provider supports prompt caching, cached_input_per_m applies to the cached part of the prompt. Cache hits show up in usage as a lower cost for the same token count.

Model ids

Ids are stable and namespaced by the model author, for example anthropic/claude-sonnet-5 or openai/gpt-5-nano. The author prefix is who trained the model. Anthropic clients that send anthropic-version receive the list in Anthropic's own shape.

Deprecated models keep working until their removal date, which is published on the model page in the catalog. Requests to a removed model return 404 with model_not_found.