DeepSeek V3.2

Overview

DeepSeek's 685B MoE featuring DeepSeek Sparse Attention for lower memory use. Holds an IMO gold-medal score and ranks #2 by volume on OpenRouter.

When to pick this model

• Frontier-class generalist tasks on a multi-GPU server
• Competition-level math and reasoning
• Replacing closed APIs with MIT-licensed weights
• High-volume production inference
• Long-context enterprise workloads

VRAM requirements by quantization

Quantization	VRAM required
Q4_K_M (recommended)	410 GB
Q5_K_M	490 GB
Q8_0	735 GB
FP16 (no quantization)	1370 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

Strengths

• IMO gold-medal reasoning quality
• DeepSeek Sparse Attention reduces memory pressure
• MIT license
• #2 by usage volume on OpenRouter

Limitations

• 410GB+ in Q4 needs a serious multi-GPU server
• Sparse attention adds inference engine complexity
• Operational overhead is significant

Architecture & training

Architecture: MoE 685B/37B active · DeepSeek Sparse Attention · MIT

Training: V3 successor with DSA for reduced memory.

Quick start

# HuggingFace : deepseek-ai/DeepSeek-V3.2 (alternative locale : ollama run deepseek-v3:671b)

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.