Skip to content

MiMo V2.6 Flash on M5 Ultra 512 GB: VRAM, speed and fit

MiMo V2.6 Flash at FP4 + FP8 runs on the M5 Ultra 512 GB with 1M of context at about 49 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on M5 Ultra 512 GB · FP4 + FP8 · Ollama / LM Studio · Chat

GPU memory
312 GB
of 512 GB
Context
1M
auto
Speed
49 tok/s
19 when full
First reply
2.8 s
2K prompt

Where the memory goes

Model
161 GB
Context
22.5 GB
Overhead
682 MB
OS
128 GB
Free
200 GB
Memory
312 GB of 512 GB
Context
1M tokens · auto
Speed
49 tok/s · 19 when full
First reply
2.8 s for a 2K prompt
Running cost
$5.06/mo at home · $287/mo on a cloud 4× A40

Validation: Validated

What MiMo V2.6 Flash on M5 Ultra 512 GB was checked against
CheckTargetResultStatus
Fits in GPU memory384 GB usable184 GB neededPass
Headroom5% free200 GB free (52%)Pass
Context for chat16K or more1MPass
Writing speed15 tok/s or more49 tok/sPass
First replyunder 3 s, 2K prompt2.8 sPass

Targets: a reply you can read along with, a first reply the use case tolerates, and 5% of memory left free. Calculated from published specs, not benchmarked.

The setup

Model
MiMo V2.6 Flash
Parameters
309B, 15B active
Precision
FP4 + FP8
Context
1M, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
M5 Ultra 512 GB
GPU can use
384 GB of 512 GB

Running cost

At home
$5.06/mo
231 W at the wall
Renting instead
$287/mo
4× A40 at $2.36/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 22, 2026.

How to run it

llama.cpp
llama-server -hf ggml-org/MiMo-V2.6-Flash-RL-GGUF
vLLM
vllm serve XiaomiMiMo/MiMo-V2.6-Flash-RL
MLX (Mac)
mlx_lm.chat --model mlx-community/MiMo-V2.6-Flash-RL-mxfp4-q8

MiMo V2.6 Flash on similar machines

All machines: MiMo V2.6 Flash VRAM requirements. Everything that runs here: what can the M5 Ultra 512 GB run?

Other models that run on M5 Ultra 512 GB

Questions

Can MiMo V2.6 Flash run on M5 Ultra 512 GB?

MiMo V2.6 Flash at FP4 + FP8 runs on the M5 Ultra 512 GB with 1M of context at about 49 tokens a second.

Report VR-0MMFQE7, calculated Nov 22, 2026 from published specs, not benchmarked.