Skip to content
OfferSign up and get $10 of credit for any model in Chat and the API. Top up from $1, with no processing fees. See free models →·Some free models are at the daily cap — refills 05:30 IST; requests auto-route to a live backup. Status →
One AI API Key
Get API key

Moonshot AI · Kimi · Flagship

Kimi K3

Discount −88%New

Moonshot AI's open-weight multimodal reasoning model, strong at large repositories, tool use and debugging.

Model id

kimi-k3
Get an API keyTry in Chat

Price per 1M tokens

Input
Regular rate $3.00now $0.353
Output
Regular rate $15.00now $5.67
Cached input
$0.3015% below input

88% below the market rate.

Added here 18 Oct 2026 — read the changelog for what changed and why.

Output is 16.1× the input rate. $1 covers about 313 prompts of 1,000 input + 500 output tokens. 1 credit = $1 of usage. All rates

Start using Kimi K3

Set ONEAIAPIKEY_API_KEY to a key from your console, then run any example. Streaming is on so long answers are not cut off by idle-connection timeouts.

curl https://api.oneaiapikey.com/v1/chat/completions \
  -H "Authorization: Bearer $ONEAIAPIKEY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k3",
    "stream": true,
    "messages": [{"role": "user", "content": "Add rate limiting to the login route, update the tests and run them."}]
  }'
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.oneaiapikey.com/v1",
    api_key=os.environ["ONEAIAPIKEY_API_KEY"],
)

stream = client.chat.completions.create(
    model="kimi-k3",
    messages=[{"role": "user", "content": "Add rate limiting to the login route, update the tests and run them."}],
    stream=True,  # recommended for long outputs
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.oneaiapikey.com/v1",
  apiKey: process.env.ONEAIAPIKEY_API_KEY,
});

const stream = await client.chat.completions.create({
  model: "kimi-k3",
  messages: [{ role: "user", content: "Add rate limiting to the login route, update the tests and run them." }],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}

More: quickstart · API reference · try it in Chat first

Benchmarks reported by the vendor

Scores for Kimi K3 as its maker published them, with the setting they state. We have not re-run them. Versions and settings differ between vendors, so compare only identical benchmark names; compare models side by side.

Tokens per day, by model

Last 7 days, UTC days

Hover, tap or drag across any day, or focus the chart and use the arrow keys, to see how every model did that day.

Real tokens served through this gateway, updated every 5 minutes. Ranked by usage, not quality.

Full ranking insights

Nearest alternative

Kimi K3 or GPT-6.1 Sol?

Kimi K3 is the only Kimi model here. GPT-6.1 Sol from OpenAI is the closest-priced flagship-tier model from another vendor. Each comparison lists the published rates and limits, then what favours each side.

Kimi K3vsGPT-6.1 Sol

Per 1M tokensKimi K3GPT-6.1 Sol
Input$0.35350% less$0.70
Output$5.6762% more$3.50
Cached input$0.303× as much$0.10
Context1.05M0% smaller1.05M

Pick Kimi K3 when

  • cost per request decides it: Kimi K3 charges input 50% less than GPT-6.1 Sol.
  • the work is agentic coding, large repositories and tool use, which this catalogue lists for Kimi K3 and not for GPT-6.1 Sol.

Pick GPT-6.1 Sol instead when

  • the job runs well enough on the lower rates: output at 38% less.
  • inputs outgrow the 1.05M-token window of Kimi K3; GPT-6.1 Sol accepts 1.05M.
  • Kimi K3 does not list Structured output and your code depends on it.
  • your workload looks like codex CLI, tool calling and long context, labels the catalogue gives GPT-6.1 Sol only.

Use cases

What Kimi K3 is listed for, and what it costs

The catalogue lists Kimi K3 for agentic coding, large repositories and tool use. Each row prices one typical request at the published rates, with the arithmetic shown. The market-rate figure shows the same request at the lowest public price elsewhere. Token counts are examples; your prompts will differ.

Model your own traffic

  1. Agentic coding

    One turn of a long agent session: 60,000 tokens of context, 50,000 of them cached from the previous turn, and 3,000 tokens of edits and tool calls back.

    Give it a clear goal and let it plan, edit and run tests across many turns.

    “Add rate limiting to the login route, update the tests and run them.”

    Same request on GPT-6.1 Sol $0.0225.

    $0.03554

    per agent turn, about 281 per $10

    $0.09 at the market rate

    0.01 × $0.353 + 0.05 × $0.30 + 0.003 × $5.67

  2. Large repositories

    A whole codebase or report in one prompt: 400,000 tokens in, a 2,000-token answer out.

    “Using the whole repository, list every place that reads the session cookie.”

    Same request on GPT-6.1 Sol $0.287.

    $0.15254

    per request, about 65 per $10

    $1.23 at the market rate

    0.4 × $0.353 + 0.002 × $5.67

  3. Tool use

    A request carrying 15 tool definitions: 5,000 tokens in, a 300-token tool call out.

    “Find open invoices older than 30 days and draft a reminder to each customer.”

    Same request on GPT-6.1 Sol $0.00455.

    $0.003466

    per tool call, about 2,885 per $10

    $0.0195 at the market rate

    0.005 × $0.353 + 0.0003 × $5.67

Overview

When Kimi K3 fits

Kimi K3 is a flagship model from Moonshot AI, available here under the exact model id “kimi-k3”. It is currently the only Kimi model we serve. Moonshot AI has not published a release date we can cite; it has been available here since 18 Oct 2026.

Moonshot AI documents tools, vision, reasoning, json mode and prompt caching for it, and it reads images as well as text.

It costs $0.353 per million input tokens and $5.67 per million output tokens, so output is 16.1× the input rate. That is 88% below the market rate of $3 input and $15 output per million tokens, the lowest public price we found at other API gateways. On the use cases below, output makes up 48% of the bill for agentic coding, 7% for large repositories and 49% for tool use, so trimming the context you send saves more than capping the answer. With a 3:1 input-to-output mix it ranks 17th lowest-cost of the 27 models on this site.

Cached input bills at $0.30 per million tokens, 15% below the standard input rate. In the agentic coding example, 50k of the 60k input tokens come from cache, which brings one agent turn from $0.03819 down to $0.03554.

It is served through the OpenAI-compatible format at https://api.oneaiapikey.com/v1, so any SDK or tool that lets you set a custom base URL, for example Codex CLI, Cursor and Cline, can use it. It is not offered on the Anthropic Messages endpoint; pick an Anthropic-format model for Claude Code.

Similar models

Same tier, other vendors

Same key, same balance. Switching is a change to the model field.

Compare side by sideAll models

Use Kimi K3 with any balance.

Kimi K3 is a paid model: any balance above $0 pays for it, the $10.00 sign-up credit included. When it runs out, add $1 or more to your wallet. Every request is billed at the rates above, with a receipt in your usage log.

Create an accountSee pricing

Kimi K3 is developed by Moonshot AI. Model names and trademarks belong to their owners. One AI API Key is an independent API gateway and is not affiliated with, endorsed by or sponsored by Moonshot AI.