Skip to content

Step 3.5 Flash on M3 Ultra 256 GB: VRAM, speed and fit

Step 3.5 Flash at 6-bit runs on the M3 Ultra 256 GB with 72K of context at about 48 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on M3 Ultra 256 GB · 6-bit (Q6_K) · Ollama / LM Studio · Chat

GPU memory
244 GB
of 256 GB
Context
72K
auto
Speed
48 tok/s
13 when full
First reply
5.5 s
2K prompt

Where the memory goes

Model
152 GB
Context
27.6 GB
Overhead
636 MB
OS
64.0 GB
Free
11.5 GB
Memory
244 GB of 256 GB
Context
72K tokens · auto
Speed
48 tok/s · 13 when full
First reply
5.5 s for a 2K prompt
Running cost
$3.55/mo at home · $287/mo on a cloud 4× A40

Validation: Validated with conditions

What Step 3.5 Flash on M3 Ultra 256 GB was checked against
CheckTargetResultStatus
Fits in GPU memory192 GB usable180 GB neededPass
Headroom5% free11.5 GB free (6%)Pass
Context for chat16K or more72KPass
Writing speed15 tok/s or more48 tok/sPass
First replyunder 3 s, 2K prompt5.5 sCheck

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
Step 3.5 Flash
Parameters
199B, 9.4B active
Precision
6-bit (Q6_K)
Context
72K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
M3 Ultra 256 GB
GPU can use
192 GB of 256 GB

Running cost

At home
$3.55/mo
162 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/Step-3.5-Flash-GGUF
vLLM
vllm serve stepfun-ai/Step-3.5-Flash
MLX (Mac)
mlx_lm.chat --model mlx-community/Step-3.5-Flash-4bit

Step 3.5 Flash on similar machines

All machines: Step 3.5 Flash VRAM requirements. Everything that runs here: what can the M3 Ultra 256 GB run?

Other models that run on M3 Ultra 256 GB

Questions

Can Step 3.5 Flash run on M3 Ultra 256 GB?

Step 3.5 Flash at 6-bit runs on the M3 Ultra 256 GB with 72K of context at about 48 tokens a second.

Report VR-0JBEV5L, calculated Nov 22, 2026 from published specs, not benchmarked.