MiMo Conversation API Integration Guide

MiMo 2.6 Pro and MiMo 2.6 Flash are Xiaomi MiMo series models. This interface uses the OpenAI Chat Completions format and charges based on input, output, and cached token hits.

¶ Models and Interface

Model	Positioning
mimo-v2.6-pro	Complex analysis, reasoning, and tool calling
mimo-v2.6-flash	Daily conversations and lightweight tasks

Enable MiMo and obtain the corresponding API credentials, then send requests to https://api.acedata.cloud/mimo/chat/completions. The base_url of the OpenAI SDK can be set to https://api.acedata.cloud/mimo. The unified /v1/chat/completions endpoint also requires the corresponding service permissions.

The initial interface scope is text Chat Completions; images, audio, video, web search, Responses, Messages, Batch, and UltraSpeed are not included in this release.

¶ Python Example

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["ACEDATACLOUD_API_KEY"],
    base_url="https://api.acedata.cloud/mimo",
)
response = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[{"role": "user", "content": "解释彩虹形成的原因。"}],
    max_completion_tokens=256,
    extra_body={"thinking": {"type": "disabled"}},
)
print(response.choices[0].message.content)
print(response.usage)

The example shows the request format; a real response example will be added after service validation.

¶ Streaming, Reasoning, and Tool Calling

Set stream=true to receive SSE; if final usage is needed, pass stream_options={"include_usage": true}. Clients should handle content, reasoning content, and tool call deltas, as well as the final usage event.

Use thinking.type=enabled to enable reasoning, and disabled to disable it. reasoning_content and final content are returned separately; when continuing tool calls, retain the reasoning content, complete tool_calls, and corresponding tool_call_id in the assistant message.

Only submit tool declarations with type=function; tool_choice supports only auto. The application is responsible for executing tools and returning the results; the model returning a tool call itself does not mean that the tool has been executed. The built-in web search service is not yet available.

¶ Billing

prompt_tokens includes cache-hit inputs; prompt_tokens_details.cached_tokens are charged at the cache price, while the remaining inputs are charged at the standard input price. completion_tokens includes reasoning output; do not double-count reasoning tokens. Streaming calls are also settled based on final actual usage.

Prices are subject to the console, and the USD cost of Credits depends on the purchased plan. Failed requests should not be considered successfully completed; if billing verification is needed, please provide the tracking identifier in the response.

¶ Error Handling

For parameter errors, check the request format; for authentication or permission errors, check the credentials and service activation status; for 429 errors, retry with backoff. A 503 may be returned when there is no healthy model capacity. If an interruption occurs after a streaming response has started, first check the results and usage already received to avoid blindly repeating the request.