API Reference

Last updated on March 11, 2025

This document contains a REST API reference for GigaChat.

Authorization

GigaChat API is available to legal entities and individual entrepreneurs after concluding an agreement. Payments are made on a postpaid basis.

GigaChat API requests are validated via OAuth 2.0 protocol using access tokens. The access token is sent in the Authorization header:

curl https://gigachat.devices.sberbank.ru/api/v1/models \
  -H 'Authorization: Bearer Token ***' \

For more information on using access tokens, see Authorization.

Description

The requests are processed under the following domain:

https://gigachat.devices.sberbank.ru/api/v1/

GET /models

Returns an array of objects containing the data for available models.

Request example:

curl https://gigachat.devices.sberbank.ru/api/v1/models \
  -H 'Authorization: Bearer Token ***' \

Response:

{
  "data": [
    {
      "id": "GigaChat:v1.2.19.2",
      "object": "model",
      "owned_by": "salutedevices"
    }
  ],
  "object": "list"
}

Models

Last updated on March 11, 2025

data

object[]

Array [

id

string

Model name

object

string

Entity type in the reply. For example, a model.

owned_by

string

Model owner

]

object

string

Entity type in the reply. For example, a list.

GET /models/{model}

Returns an object containing the description of the specified model.

Request example:

curl https://gigachat.devices.sberbank.ru/api/v1/models/GigaChat:v1.10.0 \
  -H 'Authorization: Bearer Token ***' \

Request parameters:

ModelId

Last updated on March 11, 2025

model

string

Model name

Response:

{
  "id": "GigaChat:v1.2.19.2",
  "object": "model",
  "owned_by": "salutedevices"
}

Model

Last updated on March 11, 2025

id

string

Model name

object

string

Entity type in the reply. For example, a model.

owned_by

string

Model owner

POST /chat/completions

Returns a reply of the model considering the previously sent messages.

Request example:

curl https://gigachat.devices.sberbank.ru/api/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer Token ***' \
  -d '{
  "model": "GigaChat:latest",
  "messages": [
        {
            "role": "user",
            "content": "When will the AI take over the world?"
        },
        {
            "role": "assistant",
            "content": "So far, this has not been an inevitable menace. Although the AI is rapidly developing and is able to complete tricky tasks more and more effectively, it still has limited capabilities and cannot replace a human being in all areas. Also, there are AI-related ethic and legal constratins that are considered in its development and integration."
        },
        {
            "role": "user",
            "content": "Do you think we have a chance?"
        }
    ],
  "temperature": 0.7
}'

Request parameters:

Chat

Last updated on March 11, 2025

model

string

required

Name of the model you want to get a reply from.

To select the latest model, use GigaChat:latest

messages

object[]

required

Array of messages. Send the amount of messages that the service needs to base its replies on.

For more information, see Chat History

Array [

role

string

Возможные значения: [user, assistant, system]

Message author role

content

string

Message text

]

temperature

float

Возможные значения: >= 0 и <= 2

По умолчанию: 0.87

Temperature of reply selection. The value ranges from 0 to 2. The greater the value, the more random reply the model gives.

We recommend not to use this parameter along with top_p.

top_p

float

Возможные значения: >= 0 и <= 1

По умолчанию: 0.47

This parameter is an alternative to temperature. It sets the possible array of tokens that is considered by the model. For example, if the value is 0.1, the model considers only tokens within top 10% of the probability mass.

This parameter supports values from 0 to 1.

We recommend not to use this parameter along with temperature.

n

int64

Возможные значения: >= 1 и <= 4

По умолчанию: 1

The amount of reply options to generate for each message.

stream

boolean

По умолчанию: false

Indicates that messages should be sent in parts as a stream.

The messages are sent via the SSE protocol.

A stream ends with the data: [DONE] message.

For more information, see Streaming Response Messages.

max_tokens

int64

По умолчанию: 512

The maximum ammount of tokens to use for replies.

repetition_penalty
number
По умолчанию: 1.07
The amount of repeating words:
• For values from 0 to 1, the model repeats the words that are already used
• Value 1 is a neutral value
• For values greater than 1, the model avoids repeating words
update_interval

number

По умолчанию: 0

Streaming mode parameter ("stream": "true"). This parameter sets the minimal interval between token sendings in seconds. For example, if the value is set to 1, the messages are sent every second. This means that the size of each message is larger, because a larger amount of tokens is sent during one second.

Response:

{
  "choices": [
    {
      "message": {
        "role": "assistant",
        "content": "Of course, mankind can always develop science and technologies to improve life quality and overcome challenges. It is important to continue researching the AI, find new methods of and approaches to its usage, and develop ethic principles and regulations that can help us make decisions based on our knowledge rather than emotions or prejudices."
      },
      "index": 0,
      "finish_reason": "stop"
    }
  ],
  "created": 1678878333,
  "model": "GigaChat:v1.2.19.2",
  "usage": {
    "prompt_tokens": 18,
    "completion_tokens": 68,
    "total_tokens": 86
  },
  "object": "chat.completion"
}

ChatCompletion

Last updated on March 11, 2025

choices

object[]

The array of model replies

Array [

message

object

Generated message

role

string

Возможные значения: [user, assistant, system]

Message author role

content

string

Message text

index

int32

The index of the message in the array. Begins with 0.

finish_reason

string

The cause of finishing the hypothesis. For example, stop indicates that the model has finished generating the hypothesis and given the full reply.Причина завершения гипотезы.

]

created

int64

Creation date and time in the Unix time format

model

string

Name of the model that generated the reply

usage

object

Model usage data

prompt_tokens

int32

The amount of tokens in each message

completion_tokens

int32

The amount of tokens generated by the model

total_tokens

int32

The overall amount of tokens

object

string

Name of the requested mehod

POST /tokens/count

Returns an object with tokens quantity, counted by the set model in the input array.

curl --location '{{baseUrl}}/tokens/count' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer <token>' \
--data '{
  "model": "GigaChat",
  "input": ["And the silken, sad, uncertain rustling of each purple curtain"]
}
'

Request parameters:

TokensCountBody

Last updated on March 11, 2025

model

string

required

The Model which counts tokens.

input

string[]

required

An array of strings in which tokens will be counted.

Response:

[
  {
    "object": "tokens",
    "tokens": 7,
    "characters": 36
  }
]

TokensCount

Last updated on March 11, 2025

• Array [

object

string

По умолчанию: tokens

Type of information contained in the object.

tokens

integer

Tokens quantity in a string.

characters

integer

Characters quantity in a string.

• ]

Error Codes

Error Code	Description
401	Unauthorized
400	Bad request format
404	No such Model
422	Validation error
429	Too many requests
500	Internal Server Error

See Also

• Postman Request Collection