Kimi K3 API — pricing & specs

Kimi K3 accepts text, images and video. One fixed input rate covers uncached input and one-hour cache writes; cache hits use a separate cache rate. On KeepRouter, Kimi K3 costs $5.97 per 1M input tokens and $14.92 per 1M output tokens, billed pay-as-you-go with no monthly fee. Call it through a compatible KeepRouter endpoint supported by its active route, with the model id kimi-k3.

MakerMoonshot AI
ModalityText, vision, video
Context window1,048,576 tokens
Input price$5.97 per 1M tokens
Output price$14.92 per 1M tokens
Cached input$0.3000 per 1M tokens
CapabilitiesVideo input, Vision (image input), Chat completions, Streaming where supported, Tool calling where supported
EndpointPOST /v1/chat/completions or POST /v1/messages (compatible chat routes)
Model idkimi-k3

How pricing works for Kimi K3

Kimi K3 is billed per token — $5.97 per 1M input tokens and $14.92 per 1M output tokens, with cached input at $0.3000 per 1M tokens. The published price is pay-as-you-go, with no monthly fee; actual request cost depends on measured token usage.

Calling Kimi K3 on KeepRouter

Point a compatible client at the supported KeepRouter endpoint and set the model to kimi-k3. KeepRouter preserves the client-facing request shape while handling upstream routing or translation. Use POST /v1/chat/completions or, on compatible chat routes, POST /v1/messages; streaming and tool support depend on the model's active route.

cURL

curl https://keeprouter.com/v1/chat/completions \
  -H "Authorization: Bearer $KEEPROUTER_KEY" -H "Content-Type: application/json" \
  -d '{"model":"kimi-k3","messages":[{"role":"user","content":"Hello"}]}'

Python

import os
from openai import OpenAI
client = OpenAI(base_url="https://keeprouter.com/v1", api_key=os.environ["KEEPROUTER_KEY"])
r = client.chat.completions.create(model="kimi-k3", messages=[{"role":"user","content":"Hello"}])

JavaScript

import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://keeprouter.com/v1", apiKey: process.env.KEEPROUTER_KEY });
const r = await client.chat.completions.create({ model: "kimi-k3", messages: [{ role: "user", content: "Hello" }] });

Estimate API costs

At the current KeepRouter customer price, an example workload of 1,000 total input tokens, no cached input, and 500 output tokens per request costs approximately $0.0134 per request. At 100 requests per day, that is $40.29 over 30 days. This is a usage estimate, excluding processing fees, taxes, retries and application infrastructure. Actual usage, cache hits and supported generation durations need their own checks.

Adjust quantities in the API cost calculator.

Model identity and official sources

Sources checked 2026-10-03.

Model limits and thinking

Moonshot AI documents 1,048,576 context tokens and text, image and video input. Thinking remains enabled for this model; disabling it is not a supported mode. Leave enough output budget for reasoning and the final answer.

Moonshot AI · exact model guide

Published prices and cache accounting

KeepRouter uses one fixed input rate covering uncached input and the one-hour cache-write tier. The same input rate applies to ordinary input and default five-minute writes; no second write surcharge is added. Cache hits are split out and charged at this page's cached-input rate. This fixed rate is a KeepRouter customer price, not the maker's default five-minute list rate.

Moonshot AI · cache accounting · KeepRouter · model prices

Text request verified

On October 3, 2026, a bounded text request for this exact ID returned HTTP 200, text and final token usage through KeepRouter Chat Completions. This small check does not verify media input, long-context capacity, cache hits or every tool workflow.

KeepRouter · Chat Completions reference

How to evaluate Kimi K3

Kimi K3 is listed on KeepRouter as text, vision, video under the exact id kimi-k3. Use /v1/chat/completions for the listed route; a maker's upstream features do not automatically apply to this gateway endpoint. The published context window is 1,048,576 tokens. These are limits, not a recommended request size.

First workload: Start with a text question plus one representative image, then repeat the same question with text only. Compare answer quality, latency and billed input/output tokens.

Before production: For agent use, test tool calls and structured output on this exact model route before production; an OpenAI-compatible chat endpoint alone does not prove either feature.

Agent setup paths

OpenCode

OpenCode documents a custom OpenAI-compatible provider. Set baseURL to https://keeprouter.com/v1 and list kimi-k3 as a model; validate tool behavior on the exact route. Official setup.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "keeprouter": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "KeepRouter",
      "options": {
        "baseURL": "https://keeprouter.com/v1",
        "apiKey": "{env:KEEPROUTER_API_KEY}"
      },
      "models": {
        "kimi-k3": {
          "name": "Kimi K3"
        }
      }
    }
  }
}

Continue

Continue documents provider: openai with a custom apiBase. Set apiBase to https://keeprouter.com/v1 and model to kimi-k3; validate the selected feature and endpoint. Official setup.

name: KeepRouter
version: 0.0.1
schema: v1
models:
  - name: Kimi K3
    provider: openai
    model: kimi-k3
    apiBase: https://keeprouter.com/v1
    apiKey: <YOUR_KEEPROUTER_API_KEY>

These are documented configuration paths; model-specific tool, streaming and multimodal behavior still needs a real request test.

Public model usage evidence

OpenRouter identifies the corresponding variant as moonshotai/kimi-k3. The figures below are OpenRouter token volume, not API call counts, KeepRouter traffic or market-wide usage.

App attribution is opt-in and reflects traffic through OpenRouter. It does not prove the app uses KeepRouter, endorse this gateway, or establish a model quality ranking. Tokenizers and windows can differ; do not add the weekly total to the app figures.

Source: OpenRouter model Apps observed 2026-09-28. The model Apps block does not state its measurement window.

Published examples and cases

OpenRouter's model-specific Apps view publicly attributes traffic to Hermes Agent and pi and Portkey AI. This is a public adoption signal, not a published implementation study or a KeepRouter customer claim.

Source and verification boundary

Official Kimi K3 documentation. KeepRouter's live catalog is authoritative for the customer price and enabled endpoint shown here; the maker remains authoritative for upstream model capabilities and limits.

Pricing and implementation guides

Guides

Related models

All models & pricing · Quickstart · KeepRouter vs OpenRouter · Glossary · Get an API key

Catalog facts and prices last changed .

Page content reviewed .