← All models

DeepSeek · mixture of experts · MIT

What runs DeepSeek V3.2

671B MoE with 37B active from December 2025 and still one of the most downloaded models on Hugging Face. The usual target for 512 GB Mac Studio and EPYC RAM builds.

Parameters
671B, 37B active
4-bit file
377.6 GB
Q4_K_M
Needs at 8k context
379.6 GB
context memory estimated
Longest context
163,840 tokens

Context memory: 70,272 bytes per token, an estimate from the config: MLA: (512 + 64) x 61 layers x 2 bytes. Multi-head latent attention plus DeepSeek Sparse Attention: the cache stores a 512-dim compressed latent and a 64-dim RoPE key per layer instead of per-head K/V, so the GQA formula does not apply. Card gives no per-token figure. How context memory works.

Measured on this hardware

Machines that hold the 4-bit file with 8,192 tokens of context and have published runs on the same chip or cards, fastest first. Median of single-user runs with plain decoding.

No published run on a machine that fits it.

Estimated, not measured

These fit but nobody has published a run we could match. The figure is memory bandwidth divided by the bytes read per token, scaled by the share of that ceiling measured runs reach. How estimates work.

MachineFitPriceGeneration
RTX 5090 + EPYC 9355P with 768 GB DDR5 (MoE offload)Hybrid GPU + RAMGPU + RAM offloadassumed limit~$3,763 + unpriced parts~7.2–11 tok/sestimate, part in RAM

Fits, no speed figure

Too large for

Apple Mac Studio (M5 Max, 32-core GPU), Apple Mac Studio (M5 Max, 40-core GPU), Apple Mac Studio (M5 Ultra, 64-core GPU), Apple Mac Studio (M5 Ultra, 80-core GPU), Apple Mac mini (M5 Pro, 16-core GPU), Apple Mac mini (M5 Pro, 20-core GPU), Apple MacBook Pro 14-inch (M5 Max, 32-core GPU), Apple MacBook Pro 14-inch (M5 Max, 40-core GPU), Apple MacBook Pro 16-inch (M5 Max, 32-core GPU), Apple MacBook Pro 16-inch (M5 Max, 40-core GPU), Framework Desktop (Ryzen AI Max+ 395), Framework Desktop (Ryzen AI Max 385), Framework Desktop (Ryzen AI Max+ PRO 495), GMKtec EVO-X2 (Ryzen AI Max+ 395), Beelink GTR9 Pro (Ryzen AI Max+ 395), Minisforum MS-S1 MAX (Ryzen AI Max+ 395), Corsair AI Workstation 300 (Ryzen AI Max+ 395), HP Z2 Mini G1a (Ryzen AI Max+ PRO 395), NVIDIA DGX Spark, Dell Pro Precision with GB10, Microsoft Surface RTX Spark Dev Box, Dell Pro Precision with GB300, Dell Pro Max Tower T2 with NVIDIA RTX PRO 6000 Blackwell, tiny corp tinybox red flash, 1x RTX 3090 (used), 2x RTX 3090 (used), 4x RTX 3090 (used) on EPYC 7003, open frame, 1x RTX 5090, 2x RTX 5090, 1x RTX PRO 6000 Blackwell, 2x Radeon RX 7900 XTX, 4x Instinct MI50 32GB (used) on EPYC 7003, 2x Intel Arc Pro B60, RTX 3090 (used) + 128 GB DDR5 on AM5 (MoE offload). A smaller quant may fit: try the calculator.

Files

QuantFormatSizeRepo
UD-IQ1_SGGUF171.5 GBunsloth/DeepSeek-V3.2-GGUF
UD-IQ1_MGGUF185.6 GBunsloth/DeepSeek-V3.2-GGUF
UD-IQ2_XXSGGUF202.2 GBunsloth/DeepSeek-V3.2-GGUF
UD-IQ2_MGGUF212.4 GBunsloth/DeepSeek-V3.2-GGUF
Q2_KGGUF228.5 GBunsloth/DeepSeek-V3.2-GGUF
Q2_K_LGGUF228.7 GBunsloth/DeepSeek-V3.2-GGUF
UD-Q2_K_XLGGUF229.7 GBunsloth/DeepSeek-V3.2-GGUF
UD-IQ3_XXSGGUF254.4 GBunsloth/DeepSeek-V3.2-GGUF
Q3_K_MGGUF298.2 GBunsloth/DeepSeek-V3.2-GGUF
UD-Q3_K_XLGGUF299.0 GBunsloth/DeepSeek-V3.2-GGUF
4bitMLX352.1 GBmlx-community/DeepSeek-V3.2-4bit
Q4_K_MGGUF377.6 GBunsloth/DeepSeek-V3.2-GGUF
Q5_K_MGGUF443.5 GBunsloth/DeepSeek-V3.2-GGUF
Q6_KGGUF513.5 GBunsloth/DeepSeek-V3.2-GGUF
Q8_0GGUF664.3 GBunsloth/DeepSeek-V3.2-GGUF
8bitMLX704.0 GBmlx-community/DeepSeek-V3.2-8bit

Sources