DeepSeek685B params37B active (MoE)128k contextpopular
DeepSeek V3.2 locally
DeepSeek V3.2 is an open-weight LLM from the DeepSeek family with 685B parameters. Main use: chat, RAG and general assistance. Detected minimum hardware: Mac Studio M3 Ultra (512GB) (512 GB).
Technical facts
Parameters685B
Max context128k
Q4_K_M430.6 GB
Q5_K_M526.3 GB
Q8765.5 GB
FP161531.1 GB
FamilyDeepSeek
Last sync2026-05-12
Available quantizations
GGUF weightsQ4_K_M
430.6GB
Acceptable. Good compromise when VRAM is limited.
Q5_K_M
526.3GB
Good quality. Sweet spot for size and precision.
Q8
765.5GB
Near-FP16 quality. Comfortable for production.
FP16
1531.1GB
Reference precision. Maximum quality, doubled VRAM.
Compatible GPUs
1 single-GPUGPUs that can run DeepSeek V3.2 on a single card, ranked by VRAM headroom.
Similar models
VRAM estimates: parameters x bits/8 plus margin. Real performance varies by engine, context length and batch size.
sync: 2026-05-12