Local LLM model page

DeepSeek V3.2 Exp (671B MoE)

Experimental V3.2 with DeepSeek Sparse Attention (DSA) — halves inference cost vs V3.1 on long context while keeping quality. 128K context, improved coding & tool-use. MIT licensed. Server-grade.

Find the best model for my hardwareBrowse all 183 LLMs
Parameters
671B (37B active)
Minimum RAM
512 GB
Model size
380 GB
Quantization
Q4_K_M

Can DeepSeek V3.2 Exp (671B MoE) run locally?

DeepSeek V3.2 Exp (671B MoE) is best suited for server-grade or multi-GPU systems. LocalClaw recommends Q4_K_M as the default quantization, with at least 512 GB RAM.

Search term for LM Studio or compatible runtimes: deepseek-v3.2-exp

Hugging Face repository: deepseek-ai/DeepSeek-V3.2-Exp

chatcodereasoningquality

Strengths

  • Experimental V3.2 with DeepSeek Sparse Attention (DSA) — halves inference cost vs V3.1 on long context while keeping quality. 128K context, improved coding & tool-use. MIT licensed. Server-grade.

Limitations

  • Performance depends heavily on quantization, RAM bandwidth and runtime support.

Best use cases

  • chat
  • code
  • reasoning
  • quality

Benchmarks

Speed: 2/10

Quality: 10/10

Coding: 10/10

Reasoning: 10/10

Technical details

Developer: deepseek

License: See model repository

Context window: Unknown tokens

Architecture: See model card

Released: 2025-09