DeepSeek R1 32B
by DeepSeek · deepseek-r1 family
32B
parameters
text-generation code-generation reasoning math creative-writing
DeepSeek R1 32B is a distilled reasoning model based on the Qwen 2.5 32B architecture, offering strong chain-of-thought reasoning capabilities in a size that fits on high-end consumer hardware. It provides a significant quality uplift over the 14B variant for complex reasoning tasks. This model excels at multi-step mathematical proofs, algorithmic problem solving, and analytical writing. At Q4 quantization it fits on a single 24GB GPU, making it the sweet spot for users who want powerful reasoning without requiring multi-GPU setups.
Quick Start with Ollama
ollama run 32b-q4_K_M | Creator | DeepSeek |
| Parameters | 32B |
| Architecture | transformer-decoder |
| Context Length | 128K tokens |
| License | MIT |
| Released | Jan 20, 2025 |
| Ollama | deepseek-r1:32b |
Quantization Options
| Format | File Size | VRAM Required | Quality | Ollama Tag |
|---|---|---|---|---|
| Q4_K_M recommended | 16 GB | 20.7 GB |
★
★
★
★
★
| 32b-q4_K_M |
| Q5_K_M | 18.7 GB | 23.9 GB |
★
★
★
★
★
| 32b-q5_K_M |
| Q8_0 | 28.8 GB | 34 GB |
★
★
★
★
★
| 32b-q8_0 |
Compatible Hardware for Q4_K_M
Showing compatibility for the recommended quantization (Q4_K_M, 20.7 GB VRAM).
Compatible Hardware
9 hardware
device(s) cannot run this model configuration.
Benchmark Scores
83.2
mmlu