MiMo V2.6 Pro

MiMo V2.6 Pro (checkpoint MiMo-V2.6-Pro-MOPD) is the larger model in the MiMo V2.6 family (1.02T total parameters, 42B active), served on Sarvam infrastructure as an open-weight model. It has a 1,048,576-token context window and shares its API behaviour with MiMo V2.6 Flash.

At a glance

Key capabilities

Quickstart

from sarvamai import SarvamAI
client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY")
response = client.chat.completions_v2(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "user",
            "content": "List three ways to reduce cold-start latency in a serverless API.",
        }
    ],
    max_tokens=2000,
)
print(response.choices[0].message.content)

Supported parameters

model: "mimo-v2.6-pro" on Chat Completion V2 and Responses. Full request/response schemas live on those API reference pages; this table covers what’s actually accepted for this model.

Reasoning

MiMo V2.6 Pro thinks by default. Thinking is either on or off: any effort value other than "none" (low, high, max, and so on) only switches thinking on; it doesn’t change how much the model thinks.

Turn thinking off with any of:

response = client.chat.completions_v2(
    model="mimo-v2.6-pro",
    messages=[{"role": "user", "content": "What is 17 × 23?"}],
    extra_body={"thinking": {"type": "disabled"}},
    max_tokens=2000,
)

Read the reasoning trace from choices[].message.reasoning_content on chat. On Responses, output starts with a reasoning item; read its content[0].text (summary is empty). Reasoning tokens are billed as output tokens and count against max_tokens / max_output_tokens.

Tool calling

Pass OpenAI-compatible tools. tool_choice accepts "auto", "none", "required", or a named function ({"type": "function", "function": {"name": "..."}}). The model can return several tool calls in one turn; set parallel_tool_calls: false to get at most one.

Structured output

Pass a JSON schema in response_format:

response = client.chat.completions_v2(
    model="mimo-v2.6-pro",
    messages=[{"role": "user", "content": "Give a city and its country."}],
    response_format={
        "type": "json_schema",
        "json_schema": {
            "name": "city",
            "strict": True,
            "schema": {
                "type": "object",
                "properties": {
                    "city": {"type": "string"},
                    "country": {"type": "string"},
                },
                "required": ["city", "country"],
                "additionalProperties": False,
            },
        },
    },
    max_tokens=2000,
)

Media input

MiMo V2.6 Pro accepts images, audio, and video alongside text. Send every file inline as a base64 data URI; remote http(s) URLs return 400.

import base64
def data_uri(path, mime):
    with open(path, "rb") as f:
        return f"data:{mime};base64," + base64.b64encode(f.read()).decode()
with open("question.wav", "rb") as f:
    audio_b64 = base64.b64encode(f.read()).decode()
response = client.chat.completions_v2(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "Answer the question in the audio, using the chart and the clip."},
                {"type": "image_url", "image_url": {"url": data_uri("chart.png", "image/png")}},
                {"type": "input_audio", "input_audio": {"data": audio_b64, "format": "wav"}},
                {"type": "video_url", "video_url": {"url": data_uri("clip.mp4", "video/mp4")}},
            ],
        }
    ],
    max_tokens=2000,
)

Thinking and audio: when the latest user message contains audio, the model answers without thinking by default. To think anyway, send chat_template_kwargs: {"enable_thinking": true} (through extra_body in the SDKs). Audio in earlier turns doesn’t change the default.

Limits:

• Request body up to 40 MiB, including base64 media. Base64 adds about 33%, so about 30 MB of raw media fits in one request.
• Up to 999 files of each type per request.
• Images are resized to about 8.4 megapixels.
• Video is sampled to 32 frames. Resolution isn’t capped: higher-resolution video, including 4K, is accepted and uses more tokens.
• Audio up to 10 minutes per file.

Media tokens are billed as input and reported in usage.prompt_tokens_details as image_tokens, audio_tokens, and video_tokens.

Known limitations

• Media must be inline base64; remote URLs return 400. Video works on Chat Completion V2 only.
• Plain one-colour images are often described as “white” or “black”.
• Reasoning and the final answer share the max_tokens / max_output_tokens budget.
• stop strings match the answer only, not the thinking.
• Set max_tokens explicitly. When it’s omitted, generation isn’t capped at a small default and long answers can run to tens of thousands of tokens.

Limits and errors