DeepSeek·7 sizes·Reasoning / Coding LLM
DeepSeek models: sizes & hardware to run them
The DeepSeek family spans 7 sizes from 1.5B to 671B. Each size maps to a different hardware tier — below is the approximate memory each needs at 4-bit and the device we’d start with for a private local deployment.
ReasoningCodeLong context
Sizes & hardware
| Model | Params | Context | ~VRAM @ 4-bit | Minimum device | Recommended |
|---|---|---|---|---|---|
| DeepSeek-R1 Distill 1.5B | 1.5B | 128K | ~1.5GB | NVIDIA GeForce RTX 3060 12GB | NVIDIA B200 (placeholder) |
| DeepSeek-R1 Distill 8B | 8B | 128K | ~6GB | NVIDIA GeForce RTX 3060 12GB | NVIDIA B200 (placeholder) |
| DeepSeek-R1 Distill 14B | 14B | 128K | ~10GB | NVIDIA GeForce RTX 3060 12GB | NVIDIA B200 (placeholder) |
| DeepSeek-Coder V2 (class) | 16B | 128K | ~11GB | Intel Arc A770 16GB | NVIDIA B200 (placeholder) |
| DeepSeek-R1 Distill 32B | 32B | 128K | ~20GB | NVIDIA GeForce RTX 3090 | NVIDIA B200 (placeholder) |
| DeepSeek-R1 Distill Llama 70B | 70B | 128K | ~42GB | NVIDIA RTX A6000 | NVIDIA B200 (placeholder) |
| DeepSeek-R1 671B (MoE) | 671B (≈37B active) | 128K | ~400GB | Supermicro 8x H100 SuperServer | Supermicro 8x H100 SuperServer |
Memory figures are approximate working-set estimates (weights + KV cache at modest context); treat as ±. Device picks come from our compatibility engine, best on-prem fit first.
Open each size
Reasoning
DeepSeek-R1 Distill 1.5B
Runs on almost any hardware, including CPUs and mini PCs. A reasoning model you can put on the edge.
Reasoning
DeepSeek-R1 Distill 8B
8GB+ GPUs at 4-bit — runs alongside most entry hardware. A reasoning model for a single office box.
Reasoning
DeepSeek-R1 Distill 14B
16GB+ GPUs at 4-bit. A mid-size reasoning model for analysis-heavy private agents.
Coding LLM
DeepSeek-Coder V2 (class)
The compact coder variants fit 16GB+ at 4-bit; larger MoE variants need 48GB+ or cloud.
Reasoning
DeepSeek-R1 Distill 32B
A 24GB+ card (RTX 3090/4090) at 4-bit. The best locally-runnable reasoning option for most teams.
Reasoning
DeepSeek-R1 Distill Llama 70B
Flagship tier — ~42GB at 4-bit needs a 48GB card, a 64GB+ unified-memory Mac, or multi-GPU.
Reasoning
DeepSeek-R1 671B (MoE)
Datacenter / multi-node or cloud. The full R1 is a mixture-of-experts at frontier scale — plan for a cluster.
Run DeepSeek models inside a private AI Business OS
Run your own AI agents on hardware you control — private by design, no per-seat data leaving your premises. BrainOutput helps you pick the right machine and turn it into a working AI Business OS.
Explore the AI Business OS