FitMyLLM
DeepSeek/Mixture of Experts

DeepSeekDeepSeek V3.2

DeepSeek V3.2 with 671B MoE (37B active). MIT licensed. Matches GPT-5 level on reasoning. 131K context.

chatcodingreasoningtool_usemultilingual
671B
Parameters (37B active)
128K
Context length
16
Benchmarks
17
Quantizations
Architecture
MoE
Released
2025-12-01
Layers
61
KV Heads
8
Head Dim
128
Family
deepseek

Quantization Options

Context length:
QuantBitsVRAM @ 16KQuality
IQ2_XXS2.38
203.0 GB
200.1 + 2.9 KV
low
IQ2_M2.93
249.1 GB
246.2 + 2.9 KV
low
Q2_K3.16
268.4 GB
265.5 + 2.9 KV
low
IQ3_XXS3.25
275.9 GB
273.1 + 2.9 KV
low
IQ3_XS3.5
296.9 GB
294.1 + 2.9 KV
low
Q3_K_S3.64
308.7 GB
305.8 + 2.9 KV
low
IQ3_M3.76
318.7 GB
315.9 + 2.9 KV
low
Q3_K_M4
338.8 GB
336.0 + 2.9 KV
low
Q3_K_L4.3
364.0 GB
361.2 + 2.9 KV
moderate
IQ4_XS4.46
377.4 GB
374.6 + 2.9 KV
moderate
Q4_K_S4.67
395.0 GB
392.2 + 2.9 KV
moderate
Q4_K_M4.89
413.5 GB
410.6 + 2.9 KV
good
Q5_K_S5.57
470.5 GB
467.7 + 2.9 KV
good
Q5_K_M5.7
481.4 GB
478.6 + 2.9 KV
good
Q6_K6.56
553.6 GB
550.7 + 2.9 KV
excellent
Q8_08.5
716.3 GB
713.4 + 2.9 KV
lossless
FP1616
1345.3 GB
1342.5 + 2.9 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 ~203 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 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 (16)

AIME93.1
MMLU-PRO85.0
GPQA Diamond82.4
τ²-Bench78.9
SWE-bench70.0
LiveCodeBench59.3
MATH-50059.0
AA Math59.0
BigCodeBench50.0
IFBench49.0
HLE40.8
AA Long Context39.0
SciCode38.7
AA Coding34.6
Terminal-Bench32.6
AA Intelligence32.1

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
HuggingFaceGGUF DownloadsBuild Hardware

Find the best GPU for DeepSeek V3.2

Build Hardware for DeepSeek V3.2
▸ SPEC SHEET

DeepSeek V3.2 — 671B MoE.

▸ SPECIFICATIONS
PARAMETERS
671B (37B active)
ARCHITECTURE
Mixture of Experts
CONTEXT LENGTH
128K tokens
CAPABILITIES
chat, coding, reasoning, tool_use, multilingual
RELEASE DATE
2025-12-01
PROVIDER
DeepSeek
FAMILY
deepseek
▸ VRAM REQUIREMENTS
QUANTBPWVRAMQUALITY
IQ2_XXS2.38200.1 GB65%
IQ2_M2.93246.2 GB75%
Q2_K3.16265.5 GB78%
IQ3_XXS3.25273.1 GB82%
IQ3_XS3.5294.1 GB84%
Q3_K_S3.64305.8 GB85%
IQ3_M3.76315.9 GB86%
Q3_K_M4336.0 GB88%
Q3_K_L4.3361.2 GB90%
IQ4_XS4.46374.6 GB92%
Q4_K_S4.67392.2 GB93%
Q4_K_M4.89410.6 GB94%
Q5_K_S5.57467.7 GB96%
Q5_K_M5.7478.6 GB96%
Q6_K6.56550.7 GB97%
Q8_08.5713.4 GB100%
FP16161342.5 GB100%
§ 01BENCHMARK SCORES
MMLU-PRO85.0
BigCodeBench50.0
LiveCodeBench59.3
SWE-bench70.0
AIME93.1
MATH-50059.0
GPQA Diamond82.4
HLE40.8
AA Intelligence32.1
AA Coding34.6
AA Math59.0
aa_ifbench49.0
aa_terminal_bench32.6
aa_tau278.9
aa_scicode38.7
aa_lcr39.0