Zyphra

MiMo-V2.5

Native omnimodal agentic model with 1M-token context and 15B active parameters.

Hugging Face

XiaomiMiMo/MiMo-V2.5

Copied

Get API Key

Try in Playground

About

MiMo-V2.5 is Xiaomi's natively omnimodal open-source model, a 310B-parameter Mixture-of-Experts architecture with 15B active parameters. It pairs a 729M-parameter vision transformer using hybrid window attention with a 1M-token context window, handling text and image understanding in a single unified model without a separate vision adapter.

API Usage

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.getenv("ZYPHRA_API_KEY"),
    base_url="https://api.zyphracloud.com/api/v1",
)

response = client.chat.completions.create(
    model="XiaomiMiMo/MiMo-V2.5",
    messages=[{"role": "user", "content": "Hello!"}],
)

print(response.choices[0].message.content)

Provider

Release Date

Model License

Model Architecture

Mixture-of-experts

Total parameters

Activated parameters

Context length

Input modality

Output modality

Input price

Cached input price

Output price

Xiaomi

Apr 21, 2026

MIT

Transformer

Yes

310B

15B

Text, Image

Text

$0.25

/ 1M tokens

$0.15

/ 1M tokens

$0.50

/ 1M tokens