MiMo-V2.6-Flash Web Search Tool-Calling Workflow

One-sentence takeaway

For mimo-v2.6-flash, first enable the Web Search Plugin in MiMo Console, then call the web_search tool through OpenAI Chat Completions; when real-time information is needed, use force_search: true, and use max_keyword to control the number of concurrent keywords per round and potential call costs.

Use cases

• Suitable tasks: Questions requiring real-time public information, such as news, products, and weather; Agent workflows that need to return search sources and summaries.

• Unsuitable tasks: Fixed-knowledge questions that do not require real-time information, or calls where the plugin is not enabled or the search and additional input-token costs cannot be covered.

• Applicable model version: mimo-v2.6-flash (explicitly listed in the official supported-model list); this article does not attribute example results from Pro, Pro UltraSpeed, or older models to Flash.

• Applicable client, Agent, or API: MiMo API's OpenAI Chat Completions; the documentation says that other API protocols are not currently supported.

• Recommended reasoning tier and parameters: The documentation provides no Flash-specific recommendation for the reasoning tier. Tool calling uses tool_choice: "auto"; set force_search: true when network access must be forced, and set max_keyword according to the cost ceiling.

Ready-to-use content

The following is a Flash configuration skeleton organized from the official supported-model list and Web Search tool example. The complete example on the official page sets model to mimo-v2.6-pro, so its Pro output is not treated as a Flash test result; after setting the model ID to the officially supported mimo-v2.6-flash, fill in the message content and location parameters for the business as needed.

completion = client.chat.completions.create(
    model="mimo-v2.6-flash",
    messages=[
        {"role": "user", "content": "A question requiring real-time information"}
    ],
    tools=[
        {
            "type": "web_search",
            "max_keyword": 3,
            "force_search": True,
            "limit": 1,
            # Optional: provide an approximate location for the business
            "user_location": {
                "type": "approximate",
                "country": "China",
                "region": "Hubei",
                "city": "Wuhan"
            }
        }
    ],
    tool_choice="auto"
)

Parameter meanings follow the original text: force_search forces a search; when it is not forced, the model decides whether network access is needed; max_keyword limits the maximum number of search keywords per round; user_location provides an approximate location; tool_choice: "auto" allows the model to decide whether to call the tool. limit appears in the official code, but the page does not explain its semantics; verify it against the API response and the latest reference documentation before use.

Testing or workflow steps

1. Prepare an API Key according to the official First API Call process, and use https://api.xiaomimimo.com/v1 as the base_url for the OpenAI-compatible client.

2. Go to Console → Plugin Management and enable Web Search Plugin.

3. Note that the plugin switch has an approximately 5-minute cache period; do not immediately use results to determine whether the configuration has taken effect after enabling or disabling it.

4. Send a Chat Completions request with mimo-v2.6-flash, add type: "web_search" to tools, and set tool_choice: "auto".

5. Set force_search: true for questions that must be answered based on the latest webpages; otherwise, allow the model to decide whether to search based on intent.

6. When streamed output is needed, read the search sources from the first streamed data packet; both streaming and non-streaming responses should contain search and summary content.

7. If custom functions or other tools are passed at the same time, retain the model's automatic decision-making, and record the tools actually called and their sources on the application side.

8. Adjust max_keyword according to the cost ceiling. A single search round can concurrently use multiple keywords, which may result in multiple Internet Content Plugin calls; searching webpage content also adds to the model's input tokens.

Original evidence and data

• The official page defines Web Search as a basic online search tool that helps large models obtain real-time public information such as news, products, and weather.

• The page lists four capabilities: forced search and intent recognition, returning all search sources in the first streaming packet, mixed calls with custom tools, and returning search and summary content in both streaming and non-streaming modes.

• The supported-model list explicitly includes mimo-v2.6-flash, and also lists mimo-v2.6-pro, mimo-v2.6-pro-ultraspeed, mimo-v2.5-pro, and mimo-v2.5; the latter four are recorded only as entries in the official supported-model list and were not used for the Flash conclusions in this article.

• Web Search costs consist of two parts: the search plugin usage fee and the model token fee. The page lists the search tool at ¥16 in China and $5 overseas per 1,000 calls; a single search round may make multiple concurrent calls according to max_keyword.

• The FAQ explains that the model may determine that a question does not require real-time information and therefore not search; setting force_search: true can force a search. The page also says that the plugin switch has a 5-minute cache period.

• The official example response has mimo-v2.6-pro as its model, and includes url_citation source annotations and a web_search_usage usage field. This example is not a Flash test, so this article does not treat its weather answer, number of sources, or token count as Flash data.

Applicable boundaries

• “Officially supported” does not mean that this article completed a Flash API test; the original text provides no Flash-specific request response, latency, accuracy, or token data.

• force_search: true only determines whether a search is initiated; it does not guarantee complete search results or an accurate answer. The application should still display or save source information such as url_citation and perform its own verification.

• The larger max_keyword is, the higher the number of concurrent searches and plugin fees that may be incurred in a single round; set the limit according to the question type.

• The Wuhan weather, date, location, and Pro model in the example are all part of the official example context and should not be directly copied as general business conclusions.

Reproduction notes

During collection, the official Web Search page was fully opened and read; the page update time was 2026-09-22. Reproduction requires a MiMo API Key, enabling the Web Search Plugin, and calling OpenAI Chat Completions with mimo-v2.6-flash; this article records the official workflow configuration and does not claim that an API request was completed locally or that independent measurements were taken.