DeepSeek · 32B to 80B · Mixture of experts

DeepSeek V4 Flash 0731 Spark

60.31 billion parameters in total, but only 20.36 billion are read for each token. That split is the whole point of the design: it costs the memory of a large model and the speed of a small one. Grouped-query attention keeps the cache small, so long contexts cost less here than on models of the same size.

Parameters60.31B
Active per token20.36B
Layers43
KV heads1 / 64
Native context1M
Vocabulary129k

Memory needed, by quantisation

At an 8k context window. Quality is our rough ranking of how much the compression costs you: anything at or above Q5 is hard to tell apart from the original in normal use.

Quantisation File size KV cache Total VRAM Quality What it costs you
Q8_0 59.68 GB 0.01 GB 60.54 GB 99% Lossless in practice. Use it when the memory is there.
Q6_K 46.06 GB 0.01 GB 46.92 GB 98% Very close to Q8 for two thirds of the size.
Q5_K_M 39.95 GB 0.01 GB 40.81 GB 96% The quality-first choice when Q6 will not fit.
Q4_K_M 33.91 GB 0.01 GB 34.77 GB 93% The default. Best size-to-quality ratio for local use.
IQ4_XS 29.84 GB 0.01 GB 30.7 GB 91% Importance-matrix 4-bit. Q4_K_S quality, smaller file.
Q3_K_M 27.45 GB 0.01 GB 28.31 GB 86% Degradation starts to show. A way to fit one size up.
IQ3_XXS 21.48 GB 0.01 GB 22.35 GB 79% Aggressive. Only worth it on very large models.

What a longer conversation costs

Same model at Q4_K_M, only the context window changes. This model uses sliding-window attention, so most layers stop growing past 128 tokens and the bill flattens out.

ContextKV cacheTotal VRAMFits in 8 GBFits in 12 GBFits in 24 GB
4k 0.01 GB 34.65 GB no no no
8k 0.01 GB 34.77 GB no no no
16k 0.01 GB 35.02 GB no no no
32k 0.01 GB 35.52 GB no no no
64k 0.01 GB 36.52 GB no no no
128k 0.01 GB 38.52 GB no no no

The "fits" columns allow for the roughly 0.8 GB Windows keeps for the desktop.

Which hardware runs DeepSeek V4 Flash 0731 Spark

30 of 118 consumer devices run it at a quantisation worth using, at an 8k context window. Another 26 can load it only by compressing the weights far enough to damage the model, marked with a warning below.

DeviceMemoryVerdictQuantisationUsedSpeed
RTX PRO 6000 Blackwell
NVIDIA
96 GB Runs great Q8_0 60.54 GB 72.9 t/s
Apple M3 Ultra 96GB
Apple
96 GB Runs great Q8_0 60.24 GB 31.7 t/s
Apple M3 Ultra 256GB
Apple
256 GB Runs great Q8_0 60.24 GB 31.7 t/s
Apple M3 Ultra 512GB
Apple
512 GB Runs great Q8_0 60.24 GB 31.7 t/s
Apple M1 Ultra 128GB
Apple
128 GB Runs great Q8_0 60.24 GB 31 t/s
Apple M2 Ultra 128GB
Apple
128 GB Runs great Q8_0 60.24 GB 31 t/s
Apple M2 Ultra 192GB
Apple
192 GB Runs great Q8_0 60.24 GB 31 t/s
Apple M4 Max 128GB
Apple
128 GB Runs well Q8_0 60.24 GB 21.1 t/s
RTX 6000 Ada Generation
NVIDIA
48 GB Just fits Q6_K 46.92 GB 50.6 t/s
RTX A6000
NVIDIA
48 GB Just fits Q6_K 46.92 GB 40.5 t/s
Apple M1 Ultra 64GB
Apple
64 GB Runs great Q6_K 46.62 GB 40.1 t/s
Apple M2 Ultra 64GB
Apple
64 GB Runs great Q6_K 46.62 GB 40.1 t/s
Radeon PRO W7900
AMD
48 GB Just fits Q6_K 46.92 GB 40 t/s
Apple M4 Max 64GB
Apple
64 GB Runs great Q6_K 46.62 GB 27.4 t/s
Apple M1 Max 64GB
Apple
64 GB Runs well Q6_K 46.62 GB 20.1 t/s
Apple M2 Max 64GB
Apple
64 GB Runs well Q6_K 46.62 GB 20.1 t/s
Apple M3 Max 64GB
Apple
64 GB Runs well Q6_K 46.62 GB 20.1 t/s
Apple M2 Max 96GB
Apple
96 GB Runs well Q8_0 60.24 GB 15.5 t/s
Apple M3 Max 96GB
Apple
96 GB Runs well Q8_0 60.24 GB 15.5 t/s
Apple M3 Max 128GB
Apple
128 GB Runs well Q8_0 60.24 GB 15.5 t/s
Apple M4 Max 48GB
Apple
48 GB Runs great Q4_K_M 34.47 GB 37.2 t/s
Apple M3 Max 48GB
Apple
48 GB Runs great Q4_K_M 34.47 GB 27.2 t/s
Apple M4 Pro 48GB
Apple
48 GB Runs well Q4_K_M 34.47 GB 18.6 t/s
GeForce RTX 5090
NVIDIA
32 GB Just fits IQ4_XS 30.7 GB 145.7 t/s
Apple M4 Pro 64GB
Apple
64 GB Runs well Q6_K 46.62 GB 13.7 t/s
NVIDIA DGX Spark 128GB
NVIDIA
128 GB Runs well Q8_0 60.54 GB 11.1 t/s
Jetson AGX Orin 64GB
NVIDIA
64 GB Runs well Q6_K 46.92 GB 10.8 t/s
Ryzen AI Max+ 395 64GB
AMD
64 GB Runs well Q6_K 46.92 GB 10.7 t/s
Ryzen AI Max+ 395 96GB
AMD
96 GB Runs well Q6_K 46.92 GB 10.7 t/s
Ryzen AI Max+ 395 128GB
AMD
128 GB Runs well Q6_K 46.92 GB 10.7 t/s
GeForce RTX 4090
NVIDIA
24 GB Just fits IQ3_XXS ! 22.35 GB 113.8 t/s
GeForce RTX 3090 Ti
NVIDIA
24 GB Just fits IQ3_XXS ! 22.35 GB 113.8 t/s
GeForce RTX 3090
NVIDIA
24 GB Just fits IQ3_XXS ! 22.35 GB 105.7 t/s
GeForce RTX 5090 Laptop
NVIDIA
24 GB Just fits IQ3_XXS ! 22.35 GB 101.2 t/s
Radeon RX 7900 XTX
AMD
24 GB Just fits IQ3_XXS ! 22.35 GB 95.2 t/s
RTX A5000
NVIDIA
24 GB Just fits IQ3_XXS ! 22.35 GB 86.7 t/s
Apple M4 Max 36GB
Apple
36 GB Runs great IQ3_XXS ! 22.05 GB 58.6 t/s
Apple M1 Max 32GB
Apple
32 GB Just fits IQ3_XXS ! 22.05 GB 43 t/s
Apple M2 Max 32GB
Apple
32 GB Just fits IQ3_XXS ! 22.05 GB 43 t/s
Apple M3 Max 36GB
Apple
36 GB Runs great IQ3_XXS ! 22.05 GB 43 t/s
Jetson AGX Orin 32GB
NVIDIA
32 GB Just fits IQ3_XXS ! 22.35 GB 23.1 t/s
Ryzen AI Max+ 395 32GB
AMD
32 GB Just fits IQ3_XXS ! 22.35 GB 22.9 t/s
Apple M1 Pro 32GB
Apple
32 GB Just fits IQ3_XXS ! 22.05 GB 21.5 t/s
Apple M2 Pro 32GB
Apple
32 GB Just fits IQ3_XXS ! 22.05 GB 21.5 t/s
Apple M5 32GB
Apple
32 GB Just fits IQ3_XXS ! 22.05 GB 16.4 t/s
Apple M3 Pro 36GB
Apple
36 GB Runs well IQ3_XXS ! 22.05 GB 16.1 t/s
Apple M4 32GB
Apple
32 GB Just fits IQ3_XXS ! 22.05 GB 12.9 t/s
Intel Core Ultra 9 288V 32GB
Intel
32 GB Just fits IQ3_XXS ! 22.35 GB 12.2 t/s
Ryzen AI 9 HX 370 32GB
AMD
32 GB Just fits IQ3_XXS ! 22.35 GB 11.5 t/s
Radeon RX 7900 XT
AMD
20 GB Runs great IQ2_XXS ! 15.32 GB 117.7 t/s
Apple M4 Pro 24GB
Apple
24 GB Runs great IQ2_XXS ! 15.02 GB 43.5 t/s
Apple M5 24GB
Apple
24 GB Runs well IQ2_XXS ! 15.02 GB 24.4 t/s
Apple M4 24GB
Apple
24 GB Runs well IQ2_XXS ! 15.02 GB 19.1 t/s
Ryzen AI 9 HX 370 24GB
AMD
24 GB Runs well IQ2_XXS ! 15.32 GB 17 t/s
Apple M2 24GB
Apple
24 GB Runs well IQ2_XXS ! 15.02 GB 15.9 t/s
Apple M3 24GB
Apple
24 GB Runs well IQ2_XXS ! 15.02 GB 15.9 t/s
GeForce RTX 5080
NVIDIA
16 GB Partial offload 18/43 34.77 GB 4.6 t/s
GeForce RTX 5070 Ti
NVIDIA
16 GB Partial offload 18/43 34.77 GB 4.6 t/s
GeForce RTX 4080 SUPER
NVIDIA
16 GB Partial offload 18/43 34.77 GB 4.6 t/s
GeForce RTX 4080
NVIDIA
16 GB Partial offload 18/43 34.77 GB 4.6 t/s

What to buy to run DeepSeek V4 Flash 0731 Spark

The cheapest hardware that runs it at a quantisation worth using and a speed you would not resent, at an 8k context window.

DevicePriceMemoryRuns it atSpeed
Ryzen AI Max+ 395 64GB
AMD · whole machine
$1,699 64 GB IQ4_XS 16.5 t/s Check price
Apple M4 Pro 48GB
Apple · whole machine
$1,799 48 GB Q4_K_M 18.6 t/s Check price
Apple M1 Max 64GB
Apple · whole machine
$1,900 64 GB Q6_K 20.1 t/s Check price
Ryzen AI Max+ 395 96GB
AMD · whole machine
$1,999 96 GB IQ4_XS 16.5 t/s Check price
Ryzen AI Max+ 395 128GB
AMD · whole machine
$2,199 128 GB IQ4_XS 16.5 t/s Check price
GeForce RTX 5090
NVIDIA
$2,200 32 GB IQ4_XS 145.7 t/s Check price
Apple M3 Ultra 256GB
Apple · whole machine
$5,599 256 GB Q8_0 31.7 t/s Check price
Apple M3 Ultra 512GB
Apple · whole machine
$9,499 512 GB Q8_0 31.7 t/s Check price

Indicative prices reviewed 2026-09-01; used prices are marketplace typical. Options that cost more than a cheaper one with no more memory and no more speed are hidden. Price links are Amazon affiliate links. Change the standard, the context window or the budget in the buying tool.

Where it sits in the catalogue

Source: 0xSero/deepseek-v4-flash-0731-spark on Hugging Face. Downloaded 131k times in the last month. Published 2026-08-01. Architecture figures are read from the repository's own configuration file, so they move when the model does.