DeepSeek V3.2

deepseek-ai · released 2025-12-01 · mit license

DeepSeek-V3.2 is an open-weights MoE (685B total, 37B active, Dec 2025) pairing sparse attention with strong reasoning: 82.4% GPQA Diamond, 70% SWE-bench Verified (vendor). Artificial Analysis rates the non-reasoning variant 32.

Key specs

TypeLocal open-weight
Parameters685.4B total · MoE, — active
Architecturedeepseek_v32
Context window164K tokens
Knowledge cutoff
Modalitiestext
Recommended backends
Minimum viable rig

Benchmark scores

GPQA Diamond82.4%
SWE-bench Verified70%
AIME94.17%
MMLU-Pro
BFCL v3 (tool use)
Composite score7.23
Community ratingNo reviews yet

VRAM & disk per quantization

QuantVRAMDiskRAMContext
Q4_K_M399 GB397.5 GB164K

Strengths & weaknesses

Strengths: Strong competition math (AIME 94.2); Efficient long-context via DeepSeek Sparse Attention; Open weights (MIT)

Weaknesses: AA Intelligence Index (32) trails top proprietary; Weak on long-horizon agentic tasks (APEX 7)