Skip to content

Kimi K2.6 on 4× H200 NVL: VRAM, speed and fit

Kimi K2.6 at 3-bit runs on the 4× H200 NVL with 256K of context at about 120 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on 4× H200 NVL · 3-bit (Q3_K_M) · Ollama / LM Studio · Chat

GPU memory
488 GB
of 564 GB
Context
256K
auto
Speed
120 tok/s
72 when full
First reply
0.5 s
2K prompt

Where the memory goes

Model
467 GB
Context
17.2 GB
Overhead
3.1 GB
OS
0.0 GB
Free
76.3 GB
Memory
488 GB of 564 GB
Context
256K tokens · auto
Speed
120 tok/s · 72 when full
First reply
0.5 s for a 2K prompt
Running cost
$33/mo at home · $1,162/mo on a cloud 4× MI300X

Validation: Validated

What Kimi K2.6 on 4× H200 NVL was checked against
CheckTargetResultStatus
Fits in GPU memory564 GB usable488 GB neededPass
Headroom5% free76.3 GB free (14%)Pass
Context for chat16K or more256KPass
Writing speed15 tok/s or more120 tok/sPass
First replyunder 3 s, 2K prompt0.5 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
Kimi K2.6
Parameters
1T, 32B active
Precision
3-bit (Q3_K_M)
Context
256K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
4× H200 NVL
GPU can use
564 GB of 564 GB

Running cost

At home
$33/mo
1520 W at the wall
Renting instead
$1,162/mo
4× MI300X at $9.56/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 21, 2026.

How to run it

llama.cpp
llama-server -hf unsloth/Kimi-K2.6-GGUF
vLLM
vllm serve moonshotai/Kimi-K2.6
MLX (Mac)
mlx_lm.chat --model mlx-community/Kimi-K2.6-mlx-DQ3_K_M-q8

Kimi K2.6 on similar machines

All machines: Kimi K2.6 VRAM requirements. Everything that runs here: what can the 4× H200 NVL run?

Other models that run on 4× H200 NVL

Questions

Can Kimi K2.6 run on 4× H200 NVL?

Kimi K2.6 at 3-bit runs on the 4× H200 NVL with 256K of context at about 120 tokens a second.

Report VR-0I7HE2X, calculated Nov 22, 2026 from published specs, not benchmarked.