What you need to run DeepSeek V3 671B

About DeepSeek V3 671B

DeepSeek V3 671B is a general chat model from DeepSeek. Frontier class. Needs a 512 GB Mac Studio or a server. Running it locally needs about 424.7 GB of memory, and the download is 404 GB. It is a mixture of experts model: only about 22 GB of its weights are read for each word, so it generates much faster than a dense model of the same size, but all 404 GB still has to fit in memory.

Smallest Mac that runs it

The least memory you need, by Apple chip:

• M1: does not fit
• M1 Pro: does not fit
• M1 Max: does not fit
• M1 Ultra: does not fit
• M2: does not fit
• M2 Pro: does not fit
• M2 Max: does not fit
• M2 Ultra: does not fit
• M3: does not fit
• M3 Pro: does not fit
• M3 Max: does not fit
• M3 Ultra: from 512 GB
• M4: does not fit
• M4 Pro: does not fit
• M4 Max: does not fit
• M5: does not fit
• M5 Pro: does not fit
• M5 Max: does not fit

Graphics cards

No single desktop graphics card runs it well.

Hardware to run it

As an Amazon Associate I earn from qualifying purchases.

• Smallest Mac: M3 Ultra with 512 GB. See it on Amazon
• No single desktop graphics card runs it well.
• Storage: this model is a 404 GB download, and most people end up keeping several. A fast external SSD keeps your main drive free. See SSDs on Amazon

Lighter and heavier alternatives

• gpt-oss 120B: needs about 68.8 GB, a lighter choice. OpenAI's big open model. Fast for its size on high-memory machines.
• Qwen 3.5 122B A10B: needs about 85.5 GB, a lighter choice. Near frontier quality at home, if you have 96 GB or more.

Questions