Decillion

Models

Models and chat

The model catalog, and the chat endpoint.

The catalog

Every model you can call is listed once, with its type , what it takes and produces, and its pricing in US dollars.

Model ids are creator/model , such as google/gemini-3.8-flash . A bare name ( gemini-3.8-flash ) also works when it is unambiguous.

List models

GET

/v1/models

The models this key may call, each with its prices (US dollars).

Key permission: Model catalog: Read

QUERY PARAMETERS

type

string

Only this kind: language , image , video , speech , transcription , embedding , decision or rerank .

EXAMPLE

curl

JavaScript

Python

curl https://api.decillionai.com/v1/models?type=language \
  -H "Authorization: Bearer $DECILLION_API_KEY"

Response

{
  "object": "list",
  "data": [
    {
      "id": "google/gemini-3.8-flash",
      "object": "model",
      "name": "Gemini 3.8 Flash",
      "creator": "google",
      "type": "language",
      "modalities": { "input": ["text", "image", "file"], "output": ["text"] },
      "capabilities": ["streaming", "tools", "json", "vision"],
      "context_window": 1000000,
      "pricing": [
        { "label": "input", "usd": 0.3, "per": "1M tokens" },
        { "label": "output", "usd": 1.2, "per": "1M tokens" }
      ]
    }
  ]
}

GET

/v1/models/{id}

One model, as the list shows it.

Key permission: Model catalog: Read

Create a chat completion

POST

/v1/chat/completions

Send a conversation to a model and get its reply.

Key permission: Model requests: Write

BODY PARAMETERS

model

string

required

The model to call, for example google/gemini-3.8-flash .

messages

array

required

The conversation so far: objects with a role ( system , user or assistant ) and content . Content can be text, or a list of parts that include images, audio or files.

temperature

number

0 to 2. Lower is more focused, higher more varied. Defaults to 1.

max_tokens

integer

The most tokens the reply may use. The call stops there.

stream

boolean

Send the reply as it is written, as server-sent events.

stop

string or array

Up to four sequences where the model should stop.

models

array

Other models to try, in order, if this one can't answer. See Fallback models.

EXAMPLE

curl

JavaScript

Python

curl https://api.decillionai.com/v1/chat/completions \
  -H "Authorization: Bearer $DECILLION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemini-3.8-flash",
    "messages": [
      {
        "role": "user",
        "content": "Say hello in one sentence."
      }
    ]
  }'

Response

{
  "id": "chatcmpl_9f3a1c",
  "object": "chat.completion",
  "created": 1760000000,
  "model": "google/gemini-3.8-flash",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "Hello there, it's lovely to meet you!" },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 14,
    "completion_tokens": 11,
    "total_tokens": 25,
    "cost": 0.0000186
  }
}

The response says which model answered (after any fallback), and usage.cost is what the request cost, in US dollars.

Images, audio and files in a message

A model whose modalities.input includes more than text takes them as parts of a message, in the OpenAI format:

A message with an image

{
  "role": "user",
  "content": [
    { "type": "text", "text": "What is in this picture?" },
    { "type": "image_url", "image_url": { "url": "https://example.com/cat.png" } }
  ]
}

Use input_audio for audio and file for a document you have uploaded to Storage. A model that can't take a kind of input refuses the request with a 400 .

Streaming

With stream: true the response is a stream of server-sent events. Each event carries a small piece of the reply; the last is [DONE] . Just before it, an event carries the usage for the whole call.

Event stream

data: {"choices":[{"delta":{"content":"Hello"}}]}

data: {"choices":[{"delta":{"content":" there"}}]}

data: {"choices":[],"usage":{"prompt_tokens":14,"completion_tokens":11,"cost":0.0000186}}

data: [DONE]

Fallback models

models is a list of other models to try, in order, if the one you asked for can't answer (it is down, or overloaded). Every endpoint takes it.

curl

JavaScript

Python

With a fallback

curl https://api.decillionai.com/v1/chat/completions \
  -H "Authorization: Bearer $DECILLION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemini-3.8-flash",
    "messages": [
      {
        "role": "user",
        "content": "Say hello in one sentence."
      }
    ],
    "models": [
      "openai/gpt-5-mini"
    ]
  }'

The response's model says which one answered, as do the X-Decillion-Model header and Logs. A request refused for what it asks (a bad parameter, say) is not retried on the others: they would refuse it too.

What a request costs

A language model charges for the tokens you send and the tokens you get back, each at the model's rate per million tokens:

text

cost = prompt_tokens × input_rate / 1,000,000 + completion_tokens × output_rate / 1,000,000

Other kinds of model are charged for what they produce: see Images, audio, video and more. Current rates are under Models in the console.

Organizations

Images, audio, video and more