Web search API: web_search and google_search

Web search is a server-side tool. The model decides what to search for and how many times, reads the results, and answers with source links; your code doesn’t run anything. LaoZhang API exposes it the way each vendor does: Responses for GPT and Grok, native generateContent for Gemini.

Item	GPT, Grok	Gemini
Endpoint	`POST /v1/responses`	`POST /v1beta/models/{model}:generateContent`
Tool	`{"type": "web_search"}`	`{"google_search": {}}`
Search fee	$0.01 per search	$0.014 per search
Sources	`url_citation` annotations	`groundingMetadata`

On top of the search fee, the search results the model reads are billed as input tokens, which usually adds several thousand to tens of thousands of input tokens per search. Searching takes longer than a plain request, typically from ten seconds to a minute. Set your client timeout to 300 seconds.

Supported models

Verified on October 14, 2026:

Method	Models
Responses `web_search`	`gpt-6.1-sol`, `gpt-6-sol`, `gpt-5.6-luna`, `gpt-5.4-mini`, `gpt-5.4-nano`, `grok-4.5`
Gemini `google_search`	`gemini-4-argon`, `gemini-3.8-flash`, `gemini-3.5-flash-lite`

The Chat Completions web_search_options parameter isn’t available and returns 400 Unknown parameter. If your app is built on Chat Completions, send the requests that need search through Responses.

GPT and Grok: Responses web_search

Install the SDK (python -m pip install openai) and set the LAOZHANG_API_KEY environment variable first.

import os

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["LAOZHANG_API_KEY"],
    base_url="https://api.laozhang.ai/v1",
    timeout=300.0,
)

response = client.responses.create(
    model="gpt-5.4-mini",
    input="What did the S&P 500 close at yesterday? Cite your source.",
    tools=[{"type": "web_search"}],
)
print(response.output_text)

for item in response.output:
    if item.type == "message":
        for part in item.content:
            for note in part.annotations or []:
                if note.type == "url_citation":
                    print(note.title, note.url)

response.output contains, in order:

web_search_call: one item per search, and each one is billed
message: the final answer, with url_citation sources in annotations

The same code works with model="grok-4.5". Grok may search several times for one request, which adds time and search fees.

Gemini: google_search

Install the SDK (python -m pip install google-genai):

import os

from google import genai
from google.genai import types

client = genai.Client(
    api_key=os.environ["LAOZHANG_API_KEY"],
    http_options={"base_url": "https://api.laozhang.ai", "api_version": "v1beta", "timeout": 300_000},
)

response = client.models.generate_content(
    model="gemini-3.8-flash",
    contents="What did the S&P 500 close at yesterday? Cite your source.",
    config=types.GenerateContentConfig(tools=[types.Tool(google_search=types.GoogleSearch())]),
)
print(response.text)

grounding = response.candidates[0].grounding_metadata
if grounding:
    print("Queries:", grounding.web_search_queries)
    for chunk in grounding.grounding_chunks or []:
        print(chunk.web.title, chunk.web.uri)

The search fee is charged per query in web_search_queries. grounding_chunks lists the pages the model used; each uri is a Google redirect link that opens the original page. Without the SDK, send the JSON directly:

curl -s "https://api.laozhang.ai/v1beta/models/gemini-3.8-flash:generateContent" \
  -H "x-goog-api-key: ${LAOZHANG_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{"parts": [{"text": "What did the S&P 500 close at yesterday? Cite your source."}]}],
    "tools": [{"google_search": {}}]
  }'

Gemini’s OpenAI-compatible endpoint doesn’t support this tool, so use the native format for search.

Check search fees in your logs

On the console Logs page, a request that searched shows the number of web search tool calls and their cost. Through the log query API, the same details are in web_search_billing inside the other field. Total cost = number of searches × per-search price + token cost.

FAQ

Why did the model answer without searching?

The tool lets the model search; it doesn’t force it. Models usually search when the question needs current information. To make a search more likely, ask for it in the prompt, for example “search the web and cite your sources”.

Can I limit search to certain sites?

You can ask the model in the prompt to prefer certain sites, but it may still cite others. Check the behavior with a few requests before you rely on it.

Is Nano Banana’s Google Search grounding priced the same?

Yes. When a Nano Banana image model uses Google Search grounding, each search is also $0.014; see Nano Banana 2.1.

Let the model call your own functions: Function calling
Chat Completions compared with Responses: OpenAI models
All fields of the native Gemini format: Gemini protocol
Token prices per model: Models and pricing

References

OpenAI: Web search tool
xAI: Search tools
Google: Grounding with Google Search
LaoZhang API search fees and token usage: console call logs

Get started

Image Generation & Editing APIs

Video Generation APIs

Text APIs

Multimodal & Audio APIs

Web search API: web_search and google_search

Supported models

GPT and Grok: Responses web_search

Gemini: google_search

Check search fees in your logs

FAQ

Why did the model answer without searching?

Can I limit search to certain sites?

Is Nano Banana’s Google Search grounding priced the same?

References

​Supported models

​GPT and Grok: Responses web_search

​Gemini: google_search

​Check search fees in your logs

​FAQ

​Why did the model answer without searching?

​Can I limit search to certain sites?

​Is Nano Banana’s Google Search grounding priced the same?

​Related documentation

​References

Supported models

GPT and Grok: Responses web_search

Gemini: google_search

Check search fees in your logs

FAQ

Why did the model answer without searching?

Can I limit search to certain sites?

Is Nano Banana’s Google Search grounding priced the same?

Related documentation

References