Kimi API (October 2026): Kimi K3 Pricing, How to Get an API Key, Rate Limits and Code Examples

Prices and limits below are from Kimi's international API documentation at platform.kimi.ai on October 17, 2026. We checked the request formats and error messages with test keys.

Kimi API pricing

Models and prices

Per million tokens, in USD:

• Kimi K3 is Kimi's most capable model, with 2.8 trillion parameters, native image understanding and a 1M-token context window, aimed at software engineering, knowledge work and deep reasoning.
• K2.7 Code is the dedicated coding model. The highspeed version costs twice as much and outputs about 180 tokens per second, up to 260 on short contexts.
• K2.6 takes text and images and supports both thinking and non-thinking modes.

Caching: automatic, with a write charge on K3

The Kimi API caches repeated request prefixes automatically. On K3, writing to the cache costs $3.00 per million tokens with the default 5-minute lifetime or $6.00 with a 1-hour lifetime, and reading costs $0.30. Each hit refreshes the entry's lifetime at no extra write charge. On the K2 models, the table shows a cache-hit price instead.

K3 always thinks, at maximum effort by default

Kimi K3 always runs in thinking mode, and its reasoning_effort defaults to max. That gives the best answers but also the most output tokens, billed at $15 per million. For simpler requests, set reasoning_effort to high or low:

resp = client.chat.completions.create(
    model="kimi-k3",
    messages=[{"role": "user", "content": "Summarize this paragraph: ..."}],
    reasoning_effort="low",
)

For multi-turn conversations and tool calls, Kimi asks you to send back the complete assistant message from the previous response, not just its content.

Other charges

• Batch API: 60% of the standard price, for jobs that don't need an immediate answer.
• Web search tools, billed only when results come back: Basic $0.002 per call, Pro $0.003 per call, URL fetch $0.002 per call.

Kimi vs Claude, GPT and DeepSeek

At list prices Kimi K3 costs more than Claude Sonnet 5.5 and GPT-6.1 Sol and less than Claude Opus 5.5. K2.7 Code is the budget option, priced close to DeepSeek V4 Pro at peak. Claude and OpenAI prices are from our Claude and OpenAI pricing pages (October 15, 2026), and DeepSeek's from our DeepSeek API pricing page.

How to get a Kimi API key

1. Sign up on the international Kimi API Platform.
2. Recharge at least $1. Kimi requires it before you can use the API, to prevent abuse. When your total recharge reaches $5, you get a $5 voucher.
3. Create an API key in the console.
4. Store it in an environment variable, for example export MOONSHOT_API_KEY="sk-...".

There's no ongoing free tier. Free Kimi API offers on third-party sites are those sites' own promotions, not Kimi's.

International or China platform?

They're separate platforms with their own consoles. Use the base URL of the platform where you created your key.

Rate limits depend on how much you've recharged

Kimi sets limits by cumulative recharge amount, per account:

Tier 0 is very tight: one request at a time and three a minute. If you recharge only $1 to try it and then point a coding agent at it, you'll hit 429 errors almost immediately. Recharging to $10 in total moves you to Tier 1.

Base URLs and formats

Kimi says that if your code already uses the OpenAI or Anthropic API, changing the base URL and API key is all you need.

Code examples

curl

curl https://api.moonshot.ai/v1/chat/completions \
  -H "Authorization: Bearer $MOONSHOT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "kimi-k3", "messages": [{"role": "user", "content": "Explain what an API rate limit is in one sentence."}]}'

Python with the OpenAI SDK

import os
from openai import OpenAI

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

resp = client.chat.completions.create(
    model="kimi-k3",
    messages=[{"role": "user", "content": "Explain what an API rate limit is in one sentence."}],
)
print(resp.choices[0].message.content)

Claude Code

Kimi's guide uses ~/.claude/settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.moonshot.ai/anthropic",
    "ANTHROPIC_AUTH_TOKEN": "YOUR_MOONSHOT_API_KEY",
    "ANTHROPIC_MODEL": "kimi-k3[1m]",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "kimi-k3[1m]",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "kimi-k3[1m]",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "kimi-k2.7-code",
    "ANTHROPIC_DEFAULT_FABLE_MODEL": "kimi-k3[1m]"
  }
}

The [1m] suffix tells Claude Code to use the 1M-token context window, and Claude Code strips it before sending the request. Background tasks go to the cheaper kimi-k2.7-code. For how Claude Code handles third-party endpoints in general, see How to install Claude Code.

Codex

In ~/.codex/config.toml, with your key in the KIMI_API_KEY environment variable:

model = "kimi-k3"
model_provider = "kimi"

[model_providers.kimi]
name = "Kimi"
base_url = "https://api.moonshot.ai/v1"
env_key = "KIMI_API_KEY"
wire_api = "responses"

Keep model and model_provider above the [model_providers.kimi] table, or Codex silently ignores them. See How to install Codex CLI and set up config.toml.

Errors

From Kimi's error reference. We confirmed the first two against Kimi's API with a wrong key and with no key, on the China endpoint (api.moonshot.cn), which uses the same error format. Our test machine couldn't reach the international endpoint.

Retired model names

If old code stops working, check the model name. Kimi retired kimi-latest on January 28, 2026, the kimi-k2 series (including kimi-k2-turbo-preview and kimi-k2-thinking) on May 25, 2026, and kimi-k2.5 and the whole moonshot-v1 series on August 31, 2026. Kimi recommends moving to kimi-k3.

Using Kimi through a multi-model gateway

If you call Kimi alongside Claude, GPT or Gemini, a gateway gives you one key and one bill. api.foo, the service that publishes these docs, serves kimi-k3 at its OpenAI-compatible endpoint, https://api.foo/v1. We confirmed the path with a test key. Prices are on the api.foo pricing page. If you only use Kimi, the official API gives you the Batch discount and Kimi's own web search tools.

Related

• Grok API: pricing, API key and code examples
• DeepSeek API pricing and calculator
• GLM API (Z.ai) pricing and API key
• MiniMax API pricing and API key
• Claude Code vs Codex vs Cursor vs OpenCode
• What is an OpenAI-compatible API?
• API errors: 401, 429 and how to fix them

Sources

Kimi API Platform documentation (platform.kimi.ai), checked October 17, 2026: Model list, Model inference pricing, Batch pricing, Web search pricing, Recharge and rate limiting, API overview, Errors, Kimi K3 quickstart, Claude Code and Codex guides; China platform API overview.