DeepSeek · 284B · MoE

Can you run DeepSeek-V4-Flash locally?

DeepSeek-V4-Flash is a mixture-of-experts model with 284B parameters, 13B active per token. At 4-bit or better it runs on 7 of the 120 hardware configurations listed here; the smallest is the 2x RTX PRO 6000 Blackwell (192 GB). On a 128 GB DGX Spark it runs at Q2_K, at roughly 13 tokens per second (reported).

Data as of 2 November 2026

Fastest machines for DeepSeek-V4-Flash

Every machine that runs it, fastest first. Filter by type; your machine is marked. Compare machines running it side by side

12x RTX PRO 6000 BlackwellGraphics card · 192 GB~125 tok/s our est.Q4_K_M · 177 of 190 GB
2M2 UltraMac · 192 GB · 2 variants~69 tok/s our est.Q3_K_M · 143 GB 3-bit only
3M3 UltraMac · 256 GB · 2 variants~57 tok/s our est.Q4_K_M · 177 GB
4M3 Ultra · 512 GB · 80cMac · 512 GB~33 tok/s our est.Q8_0 · 310 of 384 GB
5DGX Spark (GB10)AI mini PC · 128 GB~13 tok/s reportedQ2_K · 110 of 122 GB 2-bit only
6M5 UltraMac · 256–512 GB · 3 variantsno estimateQ4_K_M · 177 GB

Too big for 110 machines

Ryzen AI Max+ 395 · 128 GB, M2 Ultra · 128 GB · 76c, M2 Ultra · 128 GB · 60c, M1 Ultra · 128 GB · 64c, M1 Ultra · 128 GB · 48c, M5 Max · 128 GB · 40c, M4 Max · 128 GB · 40c, M3 Max · 128 GB · 40c, RTX PRO 6000 Blackwell, M5 Ultra · 96 GB · 80c, M5 Ultra · 96 GB · 64c, M3 Ultra · 96 GB · 80c, M3 Ultra · 96 GB · 60c, M2 Ultra · 96 GB, M3 Max · 96 GB · 30c, M2 Max · 96 GB · 38c, 2x RTX 5090, Ryzen AI Max+ 395 · 64 GB, M2 Ultra · 64 GB · 76c, M2 Ultra · 64 GB · 60c, M1 Ultra · 64 GB · 64c, M1 Ultra · 64 GB · 48c, M5 Max · 64 GB · 40c, M4 Max · 64 GB · 40c, M3 Max · 64 GB · 40c, M2 Max · 64 GB · 38c, M2 Max · 64 GB · 30c, M1 Max · 64 GB · 32c, M1 Max · 64 GB · 24c, M5 Pro · 64 GB · 20c, M5 Pro · 64 GB · 16c, M4 Pro · 64 GB · 20c, M4 Pro · 64 GB · 16c, 2x RTX 4090, 2x RTX 3090 Ti, 2x RTX 3090, 2x RX 7900 XTX, M5 Max · 48 GB · 40c, M4 Max · 48 GB · 40c, M3 Max · 48 GB · 40c, M5 Pro · 48 GB · 20c, M5 Pro · 48 GB · 16c, M4 Pro · 48 GB · 20c, M4 Pro · 48 GB · 16c, RTX 5090, M5 Max · 36 GB · 32c, M4 Max · 36 GB · 32c, M3 Max · 36 GB · 30c, M3 Pro · 36 GB, RTX 4090, RTX 3090 Ti, RTX 3090, RX 7900 XTX, M2 Max · 32 GB · 38c, M2 Max · 32 GB · 30c, M1 Max · 32 GB · 32c, M1 Max · 32 GB · 24c, M2 Pro · 32 GB, M1 Pro · 32 GB, M6 · 32 GB · 12c, M5 · 32 GB · 10c, M5 · 32 GB · 8c, M4 · 32 GB, RX 7900 XT, M5 Pro · 24 GB · 20c, M5 Pro · 24 GB · 16c, M4 Pro · 24 GB · 20c, M4 Pro · 24 GB · 16c, M6 · 24 GB · 12c, M6 · 24 GB · 10c, M5 · 24 GB · 10c, M5 · 24 GB · 8c, M4 · 24 GB, M3 · 24 GB, M2 · 24 GB, RTX 5080, RTX 5070 Ti, RTX 5060 Ti, RTX 4080 SUPER, RTX 4080, RTX 4070 Ti SUPER, RTX 4060 Ti, Tesla T4, Tesla P100, RX 9070 XT, RX 9060 XT, M3 Pro · 18 GB, RTX 5070, RTX 4070 Ti, RTX 4070 SUPER, RTX 4070, RTX 3060, M2 Pro · 16 GB, M1 Pro · 16 GB, M6 · 16 GB · 12c, M6 · 16 GB · 10c, M5 · 16 GB · 10c, M5 · 16 GB · 8c, M4 · 16 GB, M3 · 16 GB, M2 · 16 GB, M1 · 16 GB, RTX 5060, RTX 4060, RTX 3070, RTX 3060 Ti, RTX 2070, M3 · 8 GB, M2 · 8 GB, M1 · 8 GB. DeepSeek-V4-Flash needs about 177 GB at its smallest build.

Speeds are single-user decode estimates, not measurements. "ModelFit est." comes from ModelFit's model fitted on published runs (CC BY 4.0); "our est." is memory bandwidth divided by bytes read per token, times 0.55. Memory is what the best build needs, against what the machine can give a model. Method

DGX Spark, reported

13 to 55 tok/s

one Spark needs a 2-bit or reduced build; range across engines. Source and setup in the GPUwerk guide

About the model

Parameters: 284B
Active per token: 13B
Type: Mixture of experts
Good for: Very large fast MoE
Model card: Hugging Face
Hugging Face count: 290.94B (agrees)

Questions

Can I run DeepSeek-V4-Flash on a Mac?

Yes on a Mac with enough memory: the smallest Apple configuration listed here that runs it at 4-bit or better is the M5 Ultra · 256 GB · 80c.

How much memory does DeepSeek-V4-Flash need?

About 110 GB at the smallest build in our data, including a margin for the runtime and an 8K-token context. Longer contexts need more.

Does DeepSeek-V4-Flash run on a DGX Spark?

Yes, at Q2_K. The Spark has 128 GB of unified memory, about 121.6 GB usable.