DeepSeek/Mixture of Experts

DeepSeek V3.2-Speciale

Name: DeepSeek V3.2-Speciale
Author: DeepSeek

Reasoning-optimized variant of V3.2. Extended thinking for complex math and code tasks.

chatcodingreasoningtool_use

671B

Parameters (37B active)

128K

Context length

Benchmarks

Quantizations

Architecture

MoE

Released

2025-12-01

Layers

KV Heads

128

Head Dim

128

Family

deepseek

Quantization Options

Context length:

Quant	Bits	VRAM @ 16K	Quality
IQ2_XXS	2.38	200.9 GB 200.1 + 0.8 KV	low
IQ2_M	2.93	247.0 GB 246.2 + 0.8 KV	low
Q2_K	3.16	266.3 GB 265.5 + 0.8 KV	low
IQ3_XXS	3.25	273.9 GB 273.1 + 0.8 KV	low
IQ3_XS	3.5	294.9 GB 294.1 + 0.8 KV	low
Q3_K_S	3.64	306.6 GB 305.8 + 0.8 KV	low
IQ3_M	3.76	316.7 GB 315.9 + 0.8 KV	low
Q3_K_M	4	336.8 GB 336.0 + 0.8 KV	low
Q3_K_L	4.3	362.0 GB 361.2 + 0.8 KV	moderate
IQ4_XS	4.46	375.4 GB 374.6 + 0.8 KV	moderate
Q4_K_S	4.67	393.0 GB 392.2 + 0.8 KV	moderate
Q4_K_M	4.89	411.4 GB 410.6 + 0.8 KV	good
Q5_K_S	5.57	468.5 GB 467.7 + 0.8 KV	good
Q5_K_M	5.7	479.4 GB 478.6 + 0.8 KV	good
Q6_K	6.56	551.5 GB 550.7 + 0.8 KV	excellent
Q8_0	8.5	714.2 GB 713.4 + 0.8 KV	lossless
FP16	16	1343.3 GB 1342.5 + 0.8 KV	lossless

Select your GPU above to see speed estimates and compatibility for each quantization.

Too big for a single GPU — plan a multi-GPU deployment

Even the lightest quant needs ~201 GB. Size GPUs, replicas, TCO and scaling for a production setup. Open in Enterprise →

▸ READY TO RUN THIS?RENT BY THE HOUR

RENT A GPU AND RUN DEEPSEEK V3.2-SPECIALE NOW

Rent on Vast.ai →Or RunPod →

Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.

Community Ratings

Loading ratings...

Benchmarks (15)

AIME97.0

MATH-50096.7

AA Math96.7

LiveCodeBench89.6

MMLU-PRO85.0

GPQA Diamond82.4

SWE-bench70.0

IFBench63.9

AA Long Context59.3

BigCodeBench50.0

SciCode44.0

AA Coding37.9

Terminal-Bench34.8

AA Intelligence29.4

HLE26.1

Run this model

▸Easiest way to get started·Beginners

DOCS ↗

curl -fsSL https://ollama.com/install.sh | sh

$ollama run deepseek:671b-q4_K_M

Tag may need adjustment — check ollama.com/library/deepseek for available tags.

▸ SETUP GUIDE

Auto-setup with fitmyllm CLI

Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.

pip install fitmyllmthen run fitmyllmLearn more

Auto-detect GPULive tok/s in chatSpeed benchmarks9 inference engines

HuggingFace GGUF Downloads Build Hardware

Find the best GPU for DeepSeek V3.2-Speciale

Build Hardware for DeepSeek V3.2-Speciale

QUANT	BPW	VRAM	QUALITY
IQ2_XXS	2.38	200.1 GB	65%
IQ2_M	2.93	246.2 GB	75%
Q2_K	3.16	265.5 GB	78%
IQ3_XXS	3.25	273.1 GB	82%
IQ3_XS	3.5	294.1 GB	84%
Q3_K_S	3.64	305.8 GB	85%
IQ3_M	3.76	315.9 GB	86%
Q3_K_M	4	336.0 GB	88%
Q3_K_L	4.3	361.2 GB	90%
IQ4_XS	4.46	374.6 GB	92%
Q4_K_S	4.67	392.2 GB	93%
Q4_K_M	4.89	410.6 GB	94%
Q5_K_S	5.57	467.7 GB	96%
Q5_K_M	5.7	478.6 GB	96%
Q6_K	6.56	550.7 GB	97%
Q8_0	8.5	713.4 GB	100%
FP16	16	1342.5 GB	100%