Kimi K3: from model specs to your first API call

Source: Moonshot AI’s evaluation table, checked on the date above. These are vendor-reported results, not our own tests. Tooling, inference settings and task coverage affect outcomes; use a project-specific evaluation before choosing a model.

1 MTok = 1,000,000 tokens. Published prices exclude applicable taxes. Repeated prefixes can use automatic caching, with a default 5-minute TTL. Check the current rate card and your usage breakdown before estimating a budget.

1. 01

   Create a provider API key

   Sign in to the official platform and create an API key. Check model access and billing requirements in its console.

2. 02

   Install and configure

   Use Python 3.9+ and the OpenAI Python SDK. Set MOONSHOT_API_KEY in the shell where you will run Python; use the command for your operating system.

3. 03

   Run the Python example

   Save the Python code as kimi_example.py, then run python kimi_example.py. The request uses model kimi-k3 and the official /v1 API endpoint. On macOS / Linux, use python3 if python is not available.

4. 04

   Read the response

   The example prints the final message content. Replace the prompt with your task, and review the official guide before adding streaming or tool calls.

python -m pip install --upgrade "openai>=1.0"
$env:MOONSHOT_API_KEY="YOUR_API_KEY"

python3 -m pip install --upgrade "openai>=1.0"
export MOONSHOT_API_KEY="YOUR_API_KEY"

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["MOONSHOT_API_KEY"],
    base_url="https://api.moonshot.ai/v1",
)

completion = client.chat.completions.create(
    model="kimi-k3",
    messages=[{
        "role": "user",
        "content": "Explain how to review a pull request in five steps.",
    }],
)

print(completion.choices[0].message.content)

Run only the shell commands for your platform, replacing YOUR_API_KEY locally. Keep the key on your machine or server, outside source control and browser code. API calls can incur charges. For follow-up turns and tools, preserve the complete assistant message as required by the official guide.

Review this diff for correctness and regressions. Report only actionable issues with file paths, line numbers, a concrete failure scenario and a suggested fix. Separate confirmed issues from questions. If context is missing, name the files needed.

Diff and relevant files:
[Paste here]

Using the repository map and files below, trace the request from route to service to storage. Cite file paths and symbols for each step. Identify extension points and propose a minimal implementation plan. Do not invent files or behavior absent from the supplied material.

Task, repository map and files:
[Paste here]

Write a Blender Python script for the version specified below to create a simple parametric desk organizer using primitives. Expose dimensions as variables in meters, name the objects and add a camera and lighting. Do not delete existing scene objects or access external files. Explain how to run and inspect the script.

Blender version and dimensions:
[Fill in]

Review generated code before running it. For 3D work, validate scale, geometry and compatibility in Blender or your CAD application.

Sources checked: 2026-10-24