DeepSeek LLM (1.0, 2.0) — original base models DeepSeek Coder (1.0, 2.0) — code-focused DeepSeekMath V2 — math-focused DeepSeek V3 / V3.1 / V3.2 — general-purpose (128K context) DeepSeek-R1 / R1-Zero (671B, 37B active) — reasoning model DeepSeek V4 Pro (1.6T, 49B active) — flagship (1M context) DeepSeek V4 Flash (284B, 13B active) — efficient variant Distilled R1 variants (for local use): DeepSeek-R1-Distill-Qwen-1.5B ← (the one you're using)- DeepSeek-R1-Distill-Qwen-7B DeepSeek-R1-Distill-Llama-8B DeepSeek-R1-Distill-Qwen-14B DeepSeek-R1-Distill-Qwen-32B DeepSeek-R1-Distill-Llama-70B

https://huggingface.co/deepseek-ai)

Best HuggingFace LLMs for local run in 2026, by hardware: Low-end (8GB RAM, no GPU): - SmolLM3 3B (Hugging Face's own) — runs on laptop/phone/Raspberry Pi - Llama 3.2 3B — best all-around small model - Gemma 3 4B — great for smaller machines - Phi-4-mini 3.8B — strong reasoning for its size Mid-range (16GB RAM, ~8GB VRAM GPU): - Qwen2.5 7B — strong coding & math - Llama 3.2 7B — well-balanced chat/coding - Mistral 7B v0.3 — excellent instruction following High-end (32GB+ RAM, 24GB+ VRAM): - Llama 3.3 70B (4-bit quantized) — best general purpose - DeepSeek-R1-Distill-Qwen-32B — strong reasoning - Qwen2.5 72B (quantized) — near-frontier quality Best for coding: Qwen2.5-Coder 7B or 32B Best for reasoning: DeepSeek R1 Distill (7B/14B/32B/70B) 21.05.26

• Home
• Notebook
• Contact us

• Privacy Statement
• 

Copyleft ☠ 2026