XiaomiMiMoMiMoV2FlashForCausalLM

MiMo v2 Flash — Hardware Requirements & GPU Compatibility

ChatFunctions

MiMo V2 Flash is Xiaomi's large-scale mixture-of-experts language model, built with nearly 310 billion total parameters. Designed for fast inference despite its size, the Flash variant prioritizes throughput and responsiveness, making it well-suited for interactive chat and real-time applications. Running it locally is a serious undertaking that demands high-end multi-GPU configurations, but it brings flagship-level Chinese and English language capabilities to users who have the hardware to support it.

309.8B params · MoE16B active262K contextMITDec 202541.9K downloads756 likes35.3K quant downloads

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

MiMo v2 Flash runs on a 192 GB+ Mac or a datacenter GPU

Best local pick → Mac Pro M2 Ultra (192 GB) · Q4_K_M · 187 GB · ~37 tok/s

Macs from 96 GB (IQ2_XXS · ~57 tok/s)

MIT

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run MiMo v2 Flash

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Ways to runOther ways

each opens step-by-step

Run in cloud

Runs on B300 (288 GB) (101 GB headroom) · Q4_K_M

Generation speed: ~220 tok/s; generation speed
Cost per 1M output tokens: $8.75; per 1M output tokens

Deployon RunPodBest valuefrom $6.94/hr

$5+ bonus for new users

Compare GPUs →

Run MiMo v2 Flash

187.0 GB VRAM at 16K context · runs on one B200 192 GB

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/unsloth/MiMo-V2-Flash-GGUF:Q4_K_M
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/unsloth/MiMo-V2-Flash-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/unsloth/MiMo-V2-Flash-GGUF@Q4_K_M
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf unsloth/MiMo-V2-Flash-GGUF:Q4_K_M -c 16384 -ngl 99 --port 8080

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/unsloth/MiMo-V2-Flash-GGUF:Q4_K_M
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf unsloth/MiMo-V2-Flash-GGUF:Q4_K_M -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1. Splits the model over 8 GPUs of one server (tensor parallel).
$vllm serve XiaomiMiMo/MiMo-V2-Flash --max-model-len 16384 --tensor-parallel-size 8
Runs the original weights, not the Q4_K_M GGUF; needs ~681 GB of GPU memory.

macOS only.

MLX supports only macOS, so follow these steps for that setup.

Install mlx-lm
$pip install mlx-lm
Chat in the terminal
$mlx_lm.chat --model mlx-community/MiMo-V2-Flash-4bit
Downloads the model on first run. The MLX repo has its own quantisation; it may not match the quant selected above.
Or start a local server
Listens on http://localhost:8080.
$mlx_lm.server --model mlx-community/MiMo-V2-Flash-4bit

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$unsloth/MiMo-V2-Flash-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
B300 (288 GB)CheapestCheapest101 GB free	~220 tok/s	$8.75	Deploy on RunPodfrom $6.94/hr (opens in new tab)	288 GB · 101 GB free · ~220 tok/s · $8.75 / 1M tokens

Compare GPUs →

How Much VRAM Does MiMo v2 Flash Need?

Select a quantization level; the listed hardware adheres to it.

VRAM	+32K	+128K	File	Quality	Runs on
85.6 GB	87.1 GB	91.9 GB	101.6 GB	2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss	Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB
93.3 GB	94.8 GB	99.7 GB	84.9 GB	2-bit · i-quant extra small — Importance-weighted 2-bit, extra small	Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB
97.2 GB	98.7 GB	103.5 GB	85.2 GB	2-bit · i-quant small — Importance-weighted 2-bit, small	Runs on Instinct MI250X 128 GB
105.0 GB	106.5 GB	111.3 GB	106.7 GB	2-bit · i-quant medium — Importance-weighted 2-bit, medium	Runs on Instinct MI250X 128 GB
120.4 GB	122 GB	126.8 GB	127.5 GB	3-bit · i-quant — Importance-weighted 3-bit	Runs on Instinct MI250X 128 GB
124.3 GB	125.8 GB	130.7 GB	104.4 GB	2-bit · small K-quant — 2-bit small K-quant	Runs on Instinct MI250X 128 GB
128.2 GB	129.7 GB	134.5 GB	126.4 GB	3-bit · i-quant extra small — Importance-weighted 3-bit, extra small	Runs on H200 NVL 141 GB
132.1 GB	133.6 GB	138.4 GB	133.1 GB	3-bit · i-quant small — Importance-weighted 3-bit, small	Runs on H200 NVL 141 GB
132.1 GB	133.6 GB	138.4 GB	112.3 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Runs on H200 NVL 141 GB
135.9 GB	137.4 GB	142.3 GB	133.1 GB	3-bit small — 3-bit small quantization	Runs on H200 NVL 141 GB
139.8 GB	141.3 GB	146.1 GB	140.3 GB	3-bit · i-quant medium — Importance-weighted 3-bit, medium	Runs on H200 NVL 141 GB
151.4 GB	152.9 GB	157.8 GB	147.3 GB	3-bit medium — 3-bit medium quantization	Runs on B200 192 GB
155.3 GB	156.8 GB	161.6 GB	174.7 GB	4-bit · legacy — 4-bit legacy quantization	Runs on B200 192 GB
159.2 GB	160.7 GB	165.5 GB	146.1 GB	3-bit large — 3-bit large quantization	Runs on B200 192 GB
166.9 GB	168.4 GB	173.2 GB	164.4 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Runs on B200 192 GB
174.6 GB	176.2 GB	181 GB	175.4 GB	4-bit small — 4-bit small quantization	Runs on B200 192 GB
174.6 GB	176.2 GB	181 GB	174 GB	4-bit · i-quant non-linear — Importance-weighted 4-bit, non-linear	Runs on B200 192 GB
174.6 GB	176.2 GB	181 GB	193.3 GB	4-bit · legacy+ — 4-bit legacy quantization with offset	Runs on B200 192 GB
186.3 GB	187.8 GB	192.6 GB	186.8 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Runs on B200 192 GB
190.1 GB	191.6 GB	196.5 GB	187.6 GB	4-bit large — 4-bit large quantization	Runs on B200 192 GB
213.4 GB	214.9 GB	219.7 GB	212.5 GB	5-bit small — 5-bit small quantization	Runs on Instinct MI325X 256 GB
221.1 GB	222.6 GB	227.5 GB	219.1 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Runs on Instinct MI325X 256 GB
256.0 GB	257.5 GB	262.3 GB	253.5 GB	6-bit, very good — 6-bit quantization, very good quality	Runs on Instinct MI350X 288 GB
310.2 GB	311.7 GB	316.5 GB	328.2 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Runs on 5× H100 SXM 80 GB
619.9 GB	621.4 GB	626.3 GB	617.7 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Runs on 5× H200 SXM 141 GB

Selected Q4_K_M

4-bit medium quantization

187.0GBat 16K context

Weights: 185.9 GB
KV cache: 0.8 GB
Overhead: 0.3 GB

Server-class: ~187 GB at 16K context · one B200 192 GB

Or .

Which Hardware Can Run MiMo v2 Flash?

Q4_K_M · 186.3 GB

MiMo v2 Flash (Q4_K_M) requires 186.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 243+ GB is recommended. Using the full 262K context window can add up to 12.8 GB, bringing total usage to 199.0 GB. No consumer GPU has enough memory. 4 devices with unified memory can run MiMo v2 Flash, including Mac Studio (M3 Ultra, 512GB), Mac Studio (M3 Ultra, 256GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Consumer GPUs (up to 32 GB) and Macs up to 64 GB: too small for this model (needs ≥ 86.3 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →

How good is MiMo v2 Flash?

llmrun Score blends public benchmark boards into one number; 100 = the average open model.

Not enough data yet for an llmrun Score.

A Score needs results on enough independent boards. How the Composite Score is built

Benchmarks

LMArena Text
LMArena · Oct 2026Arena
1391.5
#57/188 open
Arena
LMArena WebDev
LMArena · Oct 2026Arena
1330.6
#44/55 open
Arena
SciCode
Epoch AILab-reported
25.9%
#65/72 open
Lab-reported
ALE-Bench
Epoch AILab-reported
738.0
#26/48 open
Lab-reported
CritPt
Epoch AILab-reported
0.0%
#84/84 open
Lab-reported

All 5 MiMo v2 Flash benchmarks

Alternatives to MiMo v2 Flash

MiMo family · 8 models →

Similar models

Model	Score	VRAM	tok/s	Why it is similar
GLM 5.3 FlashZ.ai	124.1	205.2 GB	—needs > 24 GB	same size
IQuest Q1IQuestLab	—	195.4 GB	—needs > 24 GB	same size
DeepSeek V4 Flash Vision ExpDeepSeek	—	183.3 GB	—needs > 24 GB	same size
Hy3Tencent	—	182.3 GB	—needs > 24 GB	same size
Step3StepFun	—	219.1 GB	—needs > 24 GB	same size
Motif 3Motif-Technologies	—	632.8 GB	—needs > 24 GB	same size

Specifications

Publisher: Xiaomi
Family: MiMo
Parameters: 309.8B
Active parameters: 16.4B per token
Architecture: MiMoV2FlashForCausalLM
Context Length: 262,144 tokens
Vocabulary Size: 152,576
Release Date: 2025-12-16
License: MIT

Resources

Frequently Asked Questions

How much VRAM does MiMo v2 Flash need?

MiMo v2 Flash requires 186.3 GB of VRAM at Q4_K_M, or 619.9 GB at BF16. Full 262K context adds up to 12.8 GB (199.0 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 309.8B × 4.8 bits ÷ 8 = 185.9 GB

KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 13.1 GB (at full 262K context)

VRAM usage by quantization

Q4_K_M

186.3 GB

Q4_K_M + full context

199.0 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run MiMo v2 Flash?

No — MiMo v2 Flash requires at least 85.6 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for MiMo v2 Flash?

For MiMo v2 Flash, Q4_K_M (186.3 GB) offers the best balance of quality and VRAM usage. Q4_K_L (190.1 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 85.6 GB.

VRAM requirement by quantization

IQ2_XXS

85.6 GB~53%

IQ3_XS

128.2 GB~73%

Q4_0

155.3 GB~85%

Q4_K_M ★

186.3 GB~89%

Q4_K_L

190.1 GB~90%

BF16

619.9 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run MiMo v2 Flash on a Mac?

Yes — Mac Studio (M3 Ultra, 96GB) and 20 other Macs can run MiMo v2 Flash, the smallest at IQ2_XXS. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 186.3 GB of usable unified memory (RAM minus macOS overhead), from Mac Pro M2 Ultra (192 GB).

Can I run MiMo v2 Flash locally?

Yes — MiMo v2 Flash can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is MiMo v2 Flash?

At Q4_K_M, MiMo v2 Flash can reach ~86 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)

Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.

Example: NVIDIA B200 → 9.9 GB active ÷ (8000 × 0.65) = 1.90 ms, plus 48 layers × 0.055 ms = 2.64 ms, so 1000 ÷ 4.54 ms = ~220 tok/s

Estimated speed at Q4_K_M (186.3 GB)

NVIDIA B200

~220 tok/s

NVIDIA B300

~220 tok/s

AMD Instinct MI350X

~86 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of MiMo v2 Flash?

At Q4_K_M, the download is about 185.86 GB. The full-precision BF16 version is 619.53 GB. The smallest option (IQ2_XXS) is 85.19 GB.

Which GPUs can run MiMo v2 Flash?

No single consumer GPU has enough VRAM to run MiMo v2 Flash at Q4_K_M (186.3 GB). Multi-GPU or professional hardware is required.

Which devices can run MiMo v2 Flash?

4 devices with unified memory can run MiMo v2 Flash at Q4_K_M (186.3 GB), including Mac Pro M2 Ultra (192 GB), Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio M2 Ultra (192 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.