Latest Content
MiMo V2.5 icon

MiMo V2.5

NVIDIA
Xiaomi MiMo V2.5 is a native omnimodal Mixture-of-Experts model designed for unified multimodal reasoning across text, image, video, and audio. It features 310B total parameters with 15B active, using a 48-layer architecture with 4,096 hidden size and hybrid Sliding Window and Global Attention for efficient long-context processing. The model integrates a 729M-parameter vision encoder and a dedicated audio encoder, enabling rich perception capabilities. Supporting up to a 1M token context, it excels in long-horizon reasoning and advanced agentic workflows.
TypeOmni Model
CapabilitiesText Generation, Instruction Following, Reasoning, Mathematical Reasoning+7 more
Release Date28 April, 2026
Links
Blog|HF Model Card
LicenseMIT

Inference Instructions

Deploy and run this model on NVIDIA B200 GPUs using the command below. Copy the command to get started with inference.

CONSOLE
docker run -it --rm 
 --runtime=nvidia 
 --gpus all 
 --ipc=host 
 --shm-size=128g 
 -p 8000:8000 
 -v ~/.cache/huggingface:/root/.cache/huggingface 
 -e HF_TOKEN='YOUR_HF_TOKEN' 
 vllm/vllm-openai:mimov25-cu130 
 XiaomiMiMo/MiMo-V2.5 
  --tensor-parallel-size 4 
 --enable-expert-parallel 
 --max-model-len auto 
 --max-num-batched-tokens 65536 
 --gpu-memory-utilization 0.95 
  --max-num-seqs 1024 
 --enable-auto-tool-choice 
 --reasoning-parser mimo 
 --tool-call-parser mimo 
 --generation-config vllm 
 --trust-remote-code
Note

For MiMo V2.5 support, use vllm/vllm-openai:mimov25-cu130 for CUDA 13, vllm/vllm-openai:mimov25-cu129 for CUDA 12.9, or any subsequent official vLLM release.

Model Benchmarks

Each model was tested with a fixed input size and total token volume while increasing concurrency to measure serving performance under load.

ITL vs Concurrency

Time to First Token

Throughput Scaling

Total Tokens/sec vs Avg TTFT

Vultr Cloud GPU

NVIDIA HGX B200

Deploy NVIDIA B200 on Vultr Cloud GPU

Deploy NVIDIA B200 instancesContact Sales