Kimi K3 API

Model ID: kimi-k3

Moonshot AI’s multimodal reasoning model for long-context coding, visual analysis, and complex knowledge work.

Coding · Reasoning ·

Capabilities

Context length
1048576 tokens
Max output
1048576 tokens
Default completion budget
131,072 tokens (reasoning + final answer)
Reasoning
Always on
  • Repository-wide engineering
  • Screenshot-to-UI prototyping
  • Multi-document synthesis
  • Tool-using research and knowledge assistants
  • reasoning
  • streaming
  • function_calling
  • caching
  • structured_output
  • vision

API pricing

Input
$2.7 USD per 1M tokens
Output
$13.5 USD per 1M tokens
Cached input
$0.27 USD per 1M tokens

Model introduction

Kimi K3 combines a 1,048,576-token context window with native visual understanding, tool calling, structured output, and always-on reasoning. It is designed for extended engineering tasks and knowledge workflows that combine substantial evidence with multiple steps.

Core strengths

Long-context reasoning

Reason over large codebases and connected documents.

Visual understanding

Use screenshots and visual evidence in frontend and design-to-code workflows.

Multi-step tool workflows

Maintain multi-step tool interactions with structured outputs.

Use cases

Repository-wide engineering

Connect source code, architecture notes, and test output in extended coding sessions.

Screenshot-to-UI prototyping

Reason about interface screenshots and produce text or code for implementation.

Multi-document synthesis

Synthesize findings from multiple documents with their supporting context.

Tool-using research and knowledge assistants

Use application-provided tools to gather evidence and build structured results.

Specifications and limits

Input and generated output share the available context. Maximum output is a ceiling, not a recommended request size. Use retrieval and relevant source material to keep requests focused. Kimi K3 always uses reasoning. The official default completion budget is 131,072 tokens; the maximum setting of 1,048,576 includes reasoning and the final answer, and is constrained by the full request context. Visual understanding produces text, not generated media.

Getting started

Create a BulkToken API key

Sign in to BulkToken and open API Keys to create a key. Store it in the BULKTOKEN_API_KEY environment variable on your server.

Configure the API client

Use https://bulktoken.ai/v1 as the OpenAI SDK base URL and set model to kimi-k3. Authenticate with your BulkToken API key.

Send a request and inspect usage

Call POST /v1/chat/completions. Use stream: true for incremental output, then review the request and token usage in Usage Logs. Your key must have access to the selected model.

FAQ

What model ID should I use?

The official model name is Kimi K3 and the API model ID is kimi-k3.

Can Kimi K3 generate images or videos?

No. It understands visual inputs and returns text, structured data, or tool calls. It is not an image or video generation model.

Can reasoning be disabled?

No. Kimi K3 always uses reasoning. Set an appropriate output budget because reasoning tokens contribute to usage.

What is the output limit?

The official API defaults to 131,072 completion tokens and supports a maximum setting of 1,048,576. The budget includes reasoning and the final answer; actual availability also depends on the request context and gateway.

How is usage priced?

BulkToken charges 90% of the official reference rates (10% off). Cache reads and cache writes are separate categories; writes use 5-minute or 1-hour TTL rates. A cache hit is not also a cache write. Charges depend on the usage categories reported by the route. See the Pricing section for current rates.

Can I use the OpenAI SDK?

Yes. Set base_url to https://bulktoken.ai/v1, use your BulkToken API key, and select kimi-k3.

Does function calling run tools automatically?

No. Your application executes the requested tool and sends its result back with the matching tool-call ID. Model-generated tool calls are instructions for your application, not completed actions.

Code examples

See [the Kimi K3 integration guide](/docs/kimi-k3) for Python, Node.js, and cURL examples. Install openai for the Python example.

Python

import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["BULKTOKEN_API_KEY"], base_url="https://bulktoken.ai/v1")
stream = client.chat.completions.create(
    model="kimi-k3",
    messages=[{"role": "user", "content": "Explain a hash table in one sentence."}],
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

cURL

curl https://bulktoken.ai/v1/chat/completions \
  -H "Authorization: Bearer $BULKTOKEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "kimi-k3", "messages": [{"role": "user", "content": "Hello!"}], "stream": true}'