Model

DeepSeek-R1-Distill-Qwen-32B

OPEN

DeepSeek · DeepSeek-R1 family · released Jan 20, 2025

Best of the R1 distilled dense models: Qwen2.5-32B fine-tuned on R1 traces, outperformed OpenAI o1-mini for dense-model SOTA.

ReasoningCodingVisionFunction callingTool useAgentic

1009.0

Elo · rank #211

Parameters

32B

Active params

32B (dense)

Context

128K tokens

Architecture

Dense Transformer (Qwen2.5-32B architecture, GQA)

License

Apache 2.0

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

AIMEMath72.6%#96

best: GPT-5.2 · 100.0%

BIG-Bench HardReasoning41.8%#111

best: ERNIE 4.5 300B-A47B · 94.3%

CodeforcesCoding1691#32

best: DeepSeek-V4-Pro (Think Max) · 3206

GPQA DiamondReasoning62.1%#155

best: GPT-5.6 · 94.6%

IFEvalReasoning34.6%#137

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding57.2%#103

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MATH-500Math94.3%#35

best: GPT-5 · 99.4%

MMLU-ProKnowledge46.9%#140

best: Claude Fable 5 · 91.5%

Run it locally

VRAM @ Q4

20 GB

VRAM @ FP16

64 GB

Fits on (Q4)

RTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

~40 tok/s on RTX 4090 (Q4, llama.cpp)

Quantizations

GGUF Q4 · GGUF Q4_K_M · AWQ · GPTQ · MLX

DeepSeek-R1 family

Elo progression across releases

DeepSeek-R1 671BJan 20251574.5 DeepSeek-R1-Distill-Llama-70BJan 20251384.0 DeepSeek-R1-Distill-Qwen-32BJan 20251009.0 DeepSeek-R1-ZeroJan 20251435.8 DeepSeek-R1-0528May 20251757.4

Resources

HFdeepseek-ai/DeepSeek-R1-Distill-Qwen-32B GHdeepseek-ai/DeepSeek-R1

API price weights · each benchmark row carries its own source badge (see methodology)