CanItRun Logocanitrun.

DeepSeek R1 Distill Qwen 32B

DeepSeek R1 Distill Qwen 32B needs roughly 24.6 GB VRAM at Q4_K_M quantization (75.2 GB at FP16). 63 GPUs we track can run it fully in VRAM at 8k context.

63 GPUs run this natively · 27 with CPU offload

DeepSeek32.5B params125k contextMITCommercial use ok

DeepSeek R1 Distill Qwen 32B is a 32.5B parameter dense model developed by DeepSeek. 32B distillation using Qwen architecture — great local option for hard problems.

To run DeepSeek R1 Distill Qwen 32B locally: Q4_K_M ~18-20GB — runs on 24GB GPU. The sweet spot for reasoning-focused local deployment.

MMLU-Pro 65.0%, GPQA 62.1%, Math 94.3% — reasoning quality rivaling 70B+ models.

VRAM at each quantization

Figures below assume 8k context; KV cache grows linearly as context length increases.

QuantWeightsKV cacheTotal
FP32130.0 GB2.15 GB148.0 GB
BF1665.0 GB2.15 GB75.2 GB
FP1665.0 GB2.15 GB75.2 GB
Q8_034.5 GB2.15 GB41.1 GB
Q6_K26.7 GB2.15 GB32.3 GB
Q5_K_M23.1 GB2.15 GB28.3 GB
Q4_K_Mrec19.8 GB2.15 GB24.6 GB
Q3_K_M15.6 GB2.15 GB19.9 GB
Q2_K12.4 GB2.15 GB16.3 GB
NVFP4cuda16.3 GB2.15 GB20.6 GB

KV cache is calculated at 8k context (FP16). Note that NVFP4 only runs on CUDA GPUs. Turn on TurboQuant in the calculator above for lower KV cache estimates.

Benchmarks

MMLU-Pro
65
GPQA Diamond
62.1
MATH
94.3

GPUs that run DeepSeek R1 Distill Qwen 32B natively (63)

Plus 27 GPUs that run it with CPU offload (slower)

Notes

Reasoning model — great local option for hard problems.

Hugging Face ↗Ollama ↗Released 2025-01-20

Compare DeepSeek R1 Distill Qwen 32B with other models

How to run DeepSeek R1 Distill Qwen 32B locally

816244880160320

Q4_K_M needs 24.6 GB — needs a workstation or datacenter GPU (48–80 GB).

Ollama

ollama run deepseek-r1:32b

llama.cpp

./llama-cli -m deepseek-r1-distill-qwen-32b.Q4_K_M.gguf -c 8192 -ngl 99

LM Studio: Search for 'DeepSeek R1 Distill Qwen 32B' in LM Studio. The Q4_K_M variant fits on a single 24 GB GPU -- this is the most popular local reasoning model.

Why this quantization? At 32.5B dense parameters (distilled from the full R1 into the Qwen architecture), Q4_K_M fits on a single 24 GB GPU at roughly 19 GB for weights. This is the sweet spot for local reasoning -- the distilled model retains most of the full R1's chain-of-thought capability (MATH: 94.3, GPQA: 62.1) while running on mainstream hardware. Higher quantization would push it beyond 24 GB.

Who is DeepSeek R1 Distill Qwen 32B for?

Developers and power users with a 24 GB GPU (RTX 4090, RTX 3090) who want the best reasoning model that fits on their hardware. This is widely considered the go-to local model for hard problems -- math, coding, and science tasks that smaller models struggle with. The MIT license allows unrestricted use.

Best for

  • Solving complex math problems with step-by-step reasoning (MATH: 94.3)
  • Graduate-level science and technical reasoning (GPQA: 62.1)
  • Code debugging and algorithmic problem solving with chain-of-thought
  • Research assistance requiring explicit reasoning traces
  • The best reasoning-per-dollar on a single consumer GPU

Not ideal for

  • Quick chat or simple Q&A -- the chain-of-thought reasoning adds latency even for easy questions
  • Tasks where concise output is needed -- the model tends to show its work verbosely
  • Creative writing or casual conversation where a general-purpose model would be more natural

Continue reading

vram-guides9 min
Best LLMs for 16 GB VRAM GPUs (2026)
vram-guides9 min
Best LLMs for 24 GB VRAM (2026)
vram-guides10 min
DeepSeek Family Guide: R1, V3 & Distilled Models
hardware10 min
Best GPUs for Coding Agents in 2026
tutorials12 min
llama.cpp Complete Setup Guide: Build, Configure, and Optimize
comparisons10 min
Ollama vs LM Studio vs llama.cpp: Which Tool Should You Use?
hardware9 min
Best GPUs $500-$1000 for LLMs (2026)
hardware10 min
NVIDIA RTX 40 Series for LLMs: Complete VRAM & Performance Guide
hardware10 min
NVIDIA RTX 50 Series (Blackwell) for LLMs: Complete Guide
hardware10 min
AMD Radeon for LLMs: ROCm & Vulkan Complete Guide
tutorials10 min
LLM Benchmarks Explained: MMLU, HumanEval, SWE-bench, and More

Frequently asked questions

What are the VRAM requirements for DeepSeek R1 Distill Qwen 32B?
DeepSeek R1 Distill Qwen 32B requires approximately 24.6 GB of VRAM at Q4_K_M quantization, 41.1 GB at Q8, and 75.2 GB at FP16. These numbers assume 8k context window; VRAM scales linearly with context length due to the KV cache.
How many parameters does DeepSeek R1 Distill Qwen 32B have?
DeepSeek R1 Distill Qwen 32B has 32.5 billion parameters.
Is DeepSeek R1 Distill Qwen 32B good at reasoning and math?
Yes. With a MATH score of 94.3 and MMLU-Pro of 65, DeepSeek R1 Distill Qwen 32B handles complex multi-step reasoning, analytical tasks, and problem-solving well.
Can DeepSeek R1 Distill Qwen 32B run on a 16 GB GPU?
No. At Q4_K_M, DeepSeek R1 Distill Qwen 32B needs 24.6 GB of VRAM — more than 16 GB. You will need a 48 GB GPU like the RTX 6000 Ada or a dual-GPU setup.
Can DeepSeek R1 Distill Qwen 32B run on a 24 GB GPU?
No. Even at Q4_K_M, DeepSeek R1 Distill Qwen 32B needs 24.6 GB. Consider a 48 GB card like the RTX 6000 Ada or a dual RTX 4090 setup.
What is the smallest quantization for DeepSeek R1 Distill Qwen 32B that fits in 24 GB of VRAM?
At NVFP4, DeepSeek R1 Distill Qwen 32B needs 20.6 GB — the highest-quality quantization that fits in 24 GB of VRAM.
What GPU do I need to run DeepSeek R1 Distill Qwen 32B locally?
You need a 48 GB GPU or a dual-GPU setup. At Q4_K_M, DeepSeek R1 Distill Qwen 32B needs 24.6 GB VRAM. Options: RTX 6000 Ada (48 GB), A6000 (48 GB), or 2× RTX 4090.