Kimi K3 API: a practical quickstart

1. Create a key

1. Sign up at the Moonshot/Kimi developer platform (platform.kimi.ai).
2. Add a payment method — K3 has no meaningful free API tier at launch.
3. Create an API key in the console. Store it in an environment variable, never in client code.

2. First request (Python)

from openai import OpenAI

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

resp = client.chat.completions.create(
    model="kimi-k3",
    messages=[
        {"role": "system", "content": "You are a precise coding assistant."},
        {"role": "user", "content": "Explain this stack trace: ..."},
    ],
)
print(resp.choices[0].message.content)
print(resp.usage)  # prompt_tokens / completion_tokens — your bill

3. First request (Node)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.MOONSHOT_API_KEY,
  baseURL: "https://api.moonshot.ai/v1",
});

const resp = await client.chat.completions.create({
  model: "kimi-k3",
  messages: [{ role: "user", content: "Summarize this log: ..." }],
});
console.log(resp.choices[0].message.content, resp.usage);

Exact base URL and model identifier can differ by region/account tier — confirm against the official quickstart before shipping.

4. The three settings that matter

• Context caching. If you resend a large stable prefix (system prompt, repo dump), enable Moonshot's prompt caching so repeat input bills at the discounted cached rate. This is the difference between cents and dollars per agent step — model it in the cost calculator.
• usage accounting. Log prompt_tokens and completion_tokens from every response. Reconciling them weekly against your invoice catches runaway prompts early; the token calculator helps you audit individual prompts.
• Reasoning effort. K3 launched with a single "max" reasoning level. Watch the changelog — if lighter levels ship, they will be the cheapest quality/cost dial available.

5. Streaming, images, limits

Streaming works via the standard stream: true flag. Image input follows the OpenAI content-parts format (base64 or URL). Rate limits are account-tier dependent and published in the platform console; new accounts should expect conservative defaults and request raises with usage history. For architectural context on what K3 does well — repo navigation, tool use, log/screenshot reading — see the model overview, and for the full rate table see pricing.