v2.2 · 2026-06-12
DeepSeek-V3.2
deepseek-v3-2
DeepSeek-V3.2 is a 685B reasoning model profile for local AI planning. This page records provider, license, context, quantization, hardware-fit estimates, setup hints, source links, and caveats so readers do not choose a model by name alone.
PROVIDER
DeepSeek
FAMILY
DeepSeek-V3
MODEL TYPE
reasoning
PARAMETERS
685B
MODALITIES
text
ARCHITECTURE
Mixture-of-Experts model with DeepSeek Sparse Attention
CONTEXT WINDOW
—
TRAINING TOKENS
—
RELEASE DATE
2025-12-01
SETUP HINTS
OLLAMA
ollama run deepseek-v3.2LLAMA.CPP
Use a cited GGUF build with current llama.cpp-compatible tooling.HARDWARE FIT
CPU only (no GPU)
no
8 GB RAM
unknown
16 GB RAM
unknown
32 GB RAM
unknown
8 GB VRAM
unknown
12 GB VRAM
unknown
24 GB VRAM
unknown
Apple Silicon (unified memory)
unknown
Hardware fit values are conservative local-inference estimates based on GGUF size plus runtime and KV-cache overhead. Actual requirements depend on context length, quantization, runtime, GPU offload, and other running apps.
BEST FOR
·Server-side reasoning, agent experiments, long-context research, and explaining DeepSeek’s current V3 lineage to readers clearly.
AVOID IF
·You want simple private laptop use, cheap inference, or mature local app support today yet.
CAVEATS
·Not a beginner local model.
·The V3.2 card verified here reports 685B packaged parameters; active parameter count should be manually confirmed before relying on active-parameter guidance.
·DeepSeek-V3.2-Speciale is deep-reasoning focused and does not support tool calling.
·DSA and tool-calling support may require up-to-date inference backends.
·Commercial-use status from the supplied research: Allowed under MIT license; verify deployment terms and downstream safety requirements.
·Beginner summary from supplied research: DeepSeek-V3.2 is a frontier-style, server-class DeepSeek entry for reasoning, agents, long context, and sparse attention. It should be cataloged as an advanced reference model, not a normal Ollama or LM Studio beginner choice for laptops.
SOURCE URLS
FIELD EVIDENCE
parameterSizeBhuggingface.co/deepseek-ai/DeepSeek-V3.2
activeParametersBhuggingface.co/deepseek-ai/DeepSeek-V3.2
architecturehuggingface.co/deepseek-ai/DeepSeek-V3.2
contextWindowTokenshuggingface.co/deepseek-ai/DeepSeek-V3.2
modalitieshuggingface.co/deepseek-ai/DeepSeek-V3.2
releaseDatehuggingface.co/deepseek-ai/DeepSeek-V3.2
hfGgufRepos—
q4FileSizeGb—
q5FileSizeGb—
q8FileSizeGb—
ollama—
setupHints—
supportsToolshuggingface.co/deepseek-ai/DeepSeek-V3.2
reasoningTunedhuggingface.co/deepseek-ai/DeepSeek-V3.2
embeddingModelhuggingface.co/deepseek-ai/DeepSeek-V3.2
visionModelhuggingface.co/deepseek-ai/DeepSeek-V3.2
trainingTokens—
LANGUAGES
Unknown
CAPABILITIES
tool callingreasoning tuned