KimiKimi K3 API

Name: Kimi K3 API
Price: 0.3 USD
Author: Kimi

Moonshot’s flagship Kimi model — deep reasoning with a 1M-token context.

Prompt

Images

0/8 images

Billed at this model’s published rate.

Result

Other models from Kimi

Same vendor, same API call — the differences are below.

Model	Use case	Price	Resolution	Duration	Added
Kimi K3.1 Pro	Image → Text	from $0.20 / 1M cached in	—	—	Oct 2026
Kimi K3.1 Pro Highspeed	Image → Text	from $0.20 / 1M cached in	—	—	Oct 2026

Install the SDK

TypeScript targets Node 20+ and ships with types; Python needs 3.10+ and is async. Over REST there is nothing to install: call https://api.picsart.com from any HTTP client.

bash

npm install @picsart/ai-sdk

Authenticate

Send your Picsart API key as a bearer token. The SDK does it for you once the client is created.

import { createClient } from '@picsart/ai-sdk';

const ai = createClient({
  apiKey: process.env.PICSART_API_KEY, // sent as: Authorization: Bearer <key>
  apiUrl: 'https://api.picsart.com',
});

python

import os

from picsart_ai_sdk import create_client

ai = create_client(
    api_key=os.environ["PICSART_API_KEY"],  # sent as: Authorization: Bearer <key>
    api_url="https://api.picsart.com",
)

Make a request

Call the model by its id. Over REST, one endpoint serves every model: POST /workflows/v1/models/execute with the model id and its input in the body. The outer params is the envelope every workflow takes; the inner one is this model's input, with the same names generateText() / generate_text() takes.

import { createClient } from '@picsart/ai-sdk';

const ai = createClient({
  apiKey: process.env.PICSART_API_KEY,
  apiUrl: 'https://api.picsart.com',
});

const { text } = await ai.generateText('kimi-k3', {
  prompt: "A serene mountain lake at golden hour, ultra detailed"
});

console.log(text);

python

import asyncio
import os

from picsart_ai_sdk import create_client


async def main() -> None:
    async with create_client(
        api_key=os.environ["PICSART_API_KEY"],
        api_url="https://api.picsart.com",
    ) as ai:
        res = await ai.generate_text("kimi-k3", {
            "prompt": "A serene mountain lake at golden hour, ultra detailed",
        })

        print(res.text)


asyncio.run(main())

bash

curl -X POST 'https://api.picsart.com/workflows/v1/models/execute' \
  -H 'Authorization: Bearer $PICSART_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
  "params": {
    "model": "kimi-k3",
    "params": {
      "prompt": "A serene mountain lake at golden hour, ultra detailed"
    }
  }
}'

Parameters

3 parameters, the same for the SDK and for the inner params over REST. Required ones must be supplied; the rest fall back to their defaults.

Parameter	Type	Required	Default	Details
`prompt` The message or question for the model, in plain language. With input images, what to say about them.	`text`	yes	`—`	—
`imageUrls` Images for the model to look at, as URLs.	`file (image[])`	no	`—`	—
`thinking` How much the model reasons before answering. More is slower and uses more tokens.	`enum`	no	`high`	low, high

Response

generateText() resolves to the generated text. Over REST the output is at response.result, inside the gateway's status envelope with the task id, its status and the amount charged.

json

{
  "text": "…generated text…",
  "model": "kimi-k3"
}

Overview

Alongside the unified model APIs, we expose compatibility APIs that take each vendor's original parameters exactly as the vendor defines them — nothing renamed, nothing reshaped on the way through.

That makes them the shortest path onto Picsart if you already work with the vendor directly: the request bodies you've already written keep working as they are. You keep their parameter names and defaults, and you get the vendor's full parameter set rather than the subset that is shared across every model.

The cost is that a request is written for one vendor — switching models later means rewriting it, and results come back in the vendor's own shape. When you'd rather write once and change models freely, use the unified API.

Make a request

Call the workflow with ai.apis.run() in TypeScript or Python, or hit /workflows/{workflow}/execute directly — params is passed through untouched either way.

These endpoints are addressed by workflow name rather than model id: the one in this model's header, since one model runs as one workflow. Authentication is unchanged — your Picsart API key as a bearer token (see Authentication).

import { createClient, ApiRunMode } from '@picsart/ai-sdk';

const ai = createClient({
  apiKey: process.env.PICSART_API_KEY,
  apiUrl: 'https://api.picsart.com',
});

// Calls the 'chat-completions' workflow directly — params are sent as-is.
const { result, usage } = await ai.apis.run('chat-completions', {
  model: "gpt-6.1-sol",
  messages: [
    {
      role: "developer",
      content: []
    }
  ]
}, {
  mode: ApiRunMode.SYNC,
});

console.log(result); // workflow-specific output
console.log(usage?.credits); // amount charged, in $

python

import asyncio
import os

from picsart_ai_sdk import ApiRunMode, create_client


async def main() -> None:
    async with create_client(
        api_key=os.environ["PICSART_API_KEY"],
        api_url="https://api.picsart.com",
    ) as ai:
        # Calls the 'chat-completions' workflow directly — params are sent as-is.
        response = await ai.apis.run("chat-completions", {
            "model": "gpt-6.1-sol",
            "messages": [
                {
                    "role": "developer",
                    "content": [],
                },
            ],
        }, mode=ApiRunMode.SYNC)

        print(response["result"])  # workflow-specific output
        print((response.get("usage") or {}).get("credits"))  # amount charged, in $


asyncio.run(main())

bash

curl -X POST 'https://api.picsart.com/workflows/chat-completions/execute' \
  -H 'Authorization: Bearer $PICSART_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
  "params": {
    "model": "gpt-6.1-sol",
    "messages": [
      {
        "role": "developer",
        "content": []
      }
    ]
  }
}'

Parameters

8 parameters, sent inside params. These are the vendor's own names, so they line up one-for-one with the vendor's documentation. Required ones must be supplied; the rest fall back to their defaults.

Parameter	Type	Required	Default	Details
`model` Model used for completions	`string`	yes	`—`	`gpt-6.1-solgpt-6.1-terragpt-6-astragpt-6-solgpt-6-lunagpt-5gpt-5.1gpt-5.2gpt-5.4gpt-5.4-minigpt-5.4-nanogpt-5.5gpt-5.6-solgpt-5.6-terragpt-5.6-lunagpt-5-minigpt-5-nanogpt-5-search-apigpt-4ogpt-4o-minigpt-4.1gpt-4.1-minigpt-4.1-nanoclaude-sonnet-4-5claude-sonnet-4-6claude-sonnet-5claude-sonnet-5-5claude-opus-4-8claude-opus-5claude-opus-5-5claude-haiku-4-5claude-haiku-5-5claude-fable-5claude-fable-5-1gemini-3.1-pro-previewgemini-3-flash-previewgemini-3.8-flashgemini-3.7-flashgemini-3.6-flashgemini-3.5-flashgemini-3.5-flash-litegemini-3.1-flash-litegemini-2.5-flashgemini-2.5-progemini-4-argonkimi-k3kimi-k3.1-prokimi-k3.1-pro-highspeedgrok-4.7minimax-m3minimax-m3.1-flash`
`messages` A list of messages comprising the conversation so far	`MessageParam[]`	yes	`—`	—
└`name` An optional name for the participant. Provides the model information to differentiate between participants of the same role.	`string`	no	`—`	—
└`role` The role of the messages author	`string`	yes	`—`	`developersystemuserassistanttool`
└`content` An array of content parts with a defined type. Supported options differ based on the model being used to generate the response. Can contain text, image, or audio inputs. Optional on assistant messages that carry tool_calls.	`MessageContentParam[]`	yes	`—`	—
└`type` The type of the content part.	`string`	yes	`—`	`textimage_urlfile`
└`image_url` The image content.	`object`	no	`—`	—
└`text` The text content.	`string`	no	`—`	—
└`file` The file content.	`object`	no	`—`	—
└`tool_calls` The tool calls generated by the model, echoed back from a previous assistant turn. Echo them exactly as returned — including extra_content — or Gemini 3 models reject the request.	`MessageToolCallParam[]`	no	`—`	—
└`id` The ID of the tool call, as returned by the model.	`string`	yes	`—`	—
└`type` The type of the tool call. Currently, only "function" is supported.	`string`	yes	`—`	—
└`function` The function the model called.	`object`	yes	`—`	—
└`extra_content` Vendor-specific data returned with the tool call, echoed back verbatim. Gemini 3 models reject the turn when `extra_content.google.thought_signature` is missing.	`object`	no	`—`	—
└`tool_call_id` On role "tool" messages: the ID of the tool call this message responds to.	`string`	no	`—`	—
└`reasoning_content` On assistant messages: the reasoning the model returned with that turn, echoed back verbatim. Kimi models expect it on every earlier assistant turn, tool-call turns included, to keep their reasoning across the conversation.	`string`	no	`—`	—
`temperature` What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic. We generally recommend altering this or top_p but not both.	`number`	no	`—`	0–2 e.g. `0.8`
`max_completion_tokens` An upper bound for the number of tokens that can be generated for a completion, including visible output tokens and reasoning tokens.	`number`	no	`—`	—
`reasoning_effort` Constrains effort on reasoning for reasoning models. Only applicable for o-series models. Reducing reasoning effort can result in faster responses and fewer tokens used on reasoning in a response. NOT supported together with function tools for the gpt-5.6, gpt-6 and gpt-6.1 sol/terra/luna models on this endpoint: such requests are rejected by the provider. They reason by default, so function tools are rejected on them even without reasoning_effort.	`string`	no	`—`	`lowmediumhigh`
`tool_choice` Controls which (if any) tool is called by the model. none means the model will not call any tool and instead generates a message. auto means the model can pick between generating a message or calling one or more tools. required means the model must call one or more tools.	`ToolChoiceParam \| object`	no	`—`	—
`tools` A list of tools the model may call. Currently, only functions are supported as a tool. Use this to provide a list of functions the model may generate JSON inputs for. A max of 128 functions are supported.	`FunctionTool[]`	no	`—`	—
└`name` The name of the function to be called. Must be a-z, A-Z, 0-9, or contain underscores and dashes, with a maximum length of 64.	`string`	no	`—`	—
└`description` A description of what the function does, used by the model to choose when and how to call the function.	`string`	no	`—`	—
└`parameters` The parameters the functions accepts, described as a JSON Schema object.	`object`	no	`—`	—
└`strict` Whether to enable strict schema adherence when generating the function call.	`boolean`	no	`false`	—
└`type`	`string`	yes	`function`	—
`service_tier` Specifies the processing type used for serving the request.	`string`	no	`—`	`autoflexprioritydefault`

Response

Over HTTP the output arrives inside a status envelope, at response.result. ai.apis.run() unwraps that envelope for you and resolves to { result, usage } instead. Either way the result itself is the vendor's own shape.

json

{
  "result": {
    "id": "wf_abc123",
    "created": 0,
    "model": "gpt-6.1-sol",
    "choices": [
      {
        "index": 0,
        "message": {
          "role": "…",
          "content": "…",
          "tool_calls": [],
          "reasoning_content": "…"
        },
        "finish_reason": "stop"
      }
    ],
    "usage": {
      "prompt_tokens": 0,
      "completion_tokens": 0,
      "total_tokens": 0,
      "cache_creation_input_tokens": 0,
      "cache_creation": {
        "ephemeral_5m_input_tokens": 0,
        "ephemeral_1h_input_tokens": 0
      },
      "prompt_tokens_details": {
        "cached_tokens": 0,
        "audio_tokens": 0
      },
      "completion_tokens_details": {
        "reasoning_tokens": 0,
        "audio_tokens": 0,
        "accepted_prediction_tokens": 0,
        "rejected_prediction_tokens": 0
      },
      "credits": 0
    }
  }
}

json

{
  "status": "success",
  "response": {
    "id": "wf_abc123",
    "status": "COMPLETED",
    "result": {
      "id": "wf_abc123",
      "created": 0,
      "model": "gpt-6.1-sol",
      "choices": [
        {
          "index": 0,
          "message": {},
          "finish_reason": "stop"
        }
      ],
      "usage": {
        "prompt_tokens": 0,
        "completion_tokens": 0,
        "total_tokens": 0,
        "cache_creation_input_tokens": 0,
        "cache_creation": {
          "ephemeral_5m_input_tokens": 0,
          "ephemeral_1h_input_tokens": 0
        },
        "prompt_tokens_details": {
          "cached_tokens": 0,
          "audio_tokens": 0
        },
        "completion_tokens_details": {
          "reasoning_tokens": 0,
          "audio_tokens": 0,
          "accepted_prediction_tokens": 0,
          "rejected_prediction_tokens": 0
        },
        "credits": 0
      }
    }
  }
}

Rates

What this model costs per generation, in USD. Every row is an exact rate — the columns are the settings that change it.

Tier	Rate
Cached input	$0.30 / 1M cached input tokens
Input	$3.00 / 1M input tokens
Output text	$15.00 / 1M output text tokens

KKimiKimi K3 API

Params

Result

Other models from Kimi

Install the SDK

Authenticate

Make a request

Parameters

Response

Overview

Make a request

Parameters

Response

Rates

KimiKimi K3 API