← Every modelDeepSeek · 671B total, 37B active (mixture of experts)

DeepSeek R1 671B

The full DeepSeek R1, not one of the small distilled versions that share its name. It needs over 400 GB at 4-bit, so it takes a server-class rig, such as five 96 GB cards.

ollama run deepseek-r1:671b

Needs about 406 GB of graphics memory, or unified memory on a Mac. First time? Install Ollama, then paste that into a terminal.

Download404 GB
Memory to runAbout 406 GB
Read per token22.3 GB (mixture of experts)
LicenseMIT
Context128K tokens
ReleasedJanuary 2025
Good for
  • The deepest step-by-step reasoning you can run yourself
  • Math and hard coding problems
  • Teams with a big multi-GPU server
Not the best pick for
  • Any single GPU or Mac: it won't fit
  • Quick answers: it thinks at length before replying
0 GPUs and Macs

What runs DeepSeek R1 671B

Estimated speed on each card at the smallest memory size that fits it, from memory bandwidth. As a mixture-of-experts model it reads only part of its weights per token, so it runs much faster than its size suggests.

No single card or Mac we track fits it. It needs a multi-GPU rig; see the builds.

Compare

Other models