Kimi API Internet Search Pricing

$WEB_SEARCH FEE

$0.005 each time the model searches

PLUS RESULT TOKENS

About 13,000 tokens in Moonshot's example

SEARCH API

$0.002 Basic, $0.003 Pro, $0.002 page fetch

NO SEARCH, NO FEE

If the model answers directly, no search fee

RETIREMENT

$web_search expected to end 20 October 2026

FREE

Failed and empty Search API calls

From Moonshot's tool pricing page and its "Use Web Search" guide, read on 8 October 2026. The countdown uses your device's date.

Moonshot's formula: total_tokens = prompt_tokens + search_tokens + completion_tokens. If your app stops after step 1, only the $0.005 fee is charged.

Moonshot's guide shows one real $web_search answer: 13,212 prompt tokens (13,046 of them search results) and 295 output tokens. Here is what that costs on each model.

On a small screen, swipe the table sideways to see every column.

Our arithmetic from Moonshot's example and list prices. On Kimi K3 the output would also include its thinking, so real output is often larger.

Compare the monthly cost of $web_search with calling the Search API yourself and sending only the short results.

Both include the question (200 tokens) and the answer (300 output tokens) at list prices, over 30 days. The old default of 13,000 result tokens follows Moonshot's example. The new default of 1,500 is our assumption for five titles, links and snippets; measure your own. Moonshot's guide says Pro's ranked passages usually cut tokens by about ten times compared with full pages.

Moonshot also mentions an official web-search tool as a replacement. Its K3 quickstart says that tool is being updated and is not recommended for production yet, and we found no separate price for it.

1. 1

   Find the old tool

   Search your code for $web_search and builtin_function.

2. 2

   Define your own tool

   Replace it with a normal function tool, with a name, description and a query parameter.

3. 3

   Call the Search API

   When the model calls your tool, search with /v1/tools/search and send back short results.

4. 4

   Test and compare

   Check answers and token use before 20 October.

"tools": [
  {"type": "builtin_function", "function": {"name": "$web_search"}}
]

import json, os, requests
from openai import OpenAI

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

tools = [{"type": "function", "function": {
    "name": "web_search",
    "description": "Search the web for current information",
    "parameters": {"type": "object",
                   "properties": {"query": {"type": "string"}},
                   "required": ["query"]}}}]

def web_search(query):
    r = requests.post("https://api.moonshot.ai/v1/tools/search",
                      headers={"Authorization": f"Bearer {KEY}"},
                      json={"text_query": query, "limit": 5}, timeout=30)
    r.raise_for_status()
    # keep only short fields to save tokens
    return [{"title": x["title"], "url": x["url"], "snippet": x["snippet"]}
            for x in r.json().get("search_results", [])]

messages = [{"role": "user", "content": "What is new with Kimi K3?"}]
while True:
    resp = client.chat.completions.create(
        model="kimi-k2.6", messages=messages, tools=tools)
    choice = resp.choices[0]
    if choice.finish_reason != "tool_calls":
        print(choice.message.content)
        break
    messages.append(choice.message)
    for tc in choice.message.tool_calls:
        q = json.loads(tc.function.arguments)["query"]
        messages.append({"role": "tool", "tool_call_id": tc.id,
                         "name": "web_search",
                         "content": json.dumps(web_search(q))})

The "after" example is a short sketch written for this guide from Moonshot's documented fields. We have not run it against the live API; add error handling and limits before production. Setup: Kimi API setup.

• Search API calls that fail: bad requests, rate-limit errors, server errors and timeouts.
• Searches that return no results.
• Page fetches of invalid or blocked addresses, or pages with no readable content.
• With $web_search: no search fee when the model answers without searching.

From Moonshot's tool pricing page. Model tokens are always charged at normal prices: Kimi API pricing.