DeepSeek V3.2
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.
By DeepSeek · China
Updated 2026-09-15
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.
In practice, DeepSeek V3.2 is server-class even at Q4_K_M (410 GB). Stepping up to Q8_0 nearly doubles the footprint to 735 GB, and unquantized FP16 weights take 1370 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.
Without a GPU, DeepSeek V3.2 needs roughly 512 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 1 tokens/sec on entry-level GPUs, on the order of 5 tokens/sec on a mid-range card, and up to 15 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.
What hardware do you need
The table below matches DeepSeek V3.2 to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.
| GPU memory | Example cards | Best fit for DeepSeek V3.2 |
|---|---|---|
| 8 GB | RTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GB | Does not fit — needs 410 GB at Q4_K_M |
| 12 GB | RTX 5070, RTX 5070 Ti Laptop, RTX 4080 Laptop | Does not fit — needs 410 GB at Q4_K_M |
| 16 GB | RTX 5080, RTX 4080 Super, Radeon RX 9070 XT | Does not fit — needs 410 GB at Q4_K_M |
| 24 GB | RTX 4090, Radeon RX 7900 XTX, RTX 5090 Laptop | Does not fit — needs 410 GB at Q4_K_M |
| 32 GB | RTX 5090 | Does not fit — needs 410 GB at Q4_K_M |
Which hardware should you buy to run DeepSeek V3.2?
To run DeepSeek V3.2 locally at Q4, you need ~410 GB for Q4 weights alone. Hardware option to compare: BOSGAME M5 128GB / 2TB (Ryzen AI Max+ 395). This model exceeds the practical GPU memory of this mini PC. Choose a smaller model or larger infrastructure.
As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.
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
Typical workloads
In our catalog grid, DeepSeek V3.2 is filed under Frontier Chat, Code, Reasoning — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: everyday chat, drafting and summarization.
The 125k-token context window covers long chats and mid-sized documents, though very large retrieval workloads will need chunking. The MIT license is permissive, so shipping it inside a commercial product raises no special legal questions.
Architecture & training
Architecture: MoE 685B/37B active · DeepSeek Sparse Attention · MIT
Training: V3 successor with DSA for reduced memory.
A frontier-grade MIT-licensed MoE if you can run a multi-GPU cluster.
Quick start
Install the runtime for your system: Windows, macOS or Linux. Check the exact model tag or GGUF quantization below; catalog IDs are not necessarily Ollama tags.
Start at 4096 tokens of context, then use ollama ps to check GPU/CPU placement. A default download may use a different quantization from the configurator’s memory estimate. Keep the free setup working before considering a kit.
# 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.
Your private ChatGPT, free, on your own machine in an hour — LM Studio, Ollama, Open WebUI, your documents, no cloud.
- Lifetime online access
- PDF + files
- 30-day refund
Similar models worth comparing
Frequently asked questions
How much VRAM does DeepSeek V3.2 need?
At the recommended Q4_K_M quantization, DeepSeek V3.2 needs about 410 GB of VRAM. Q8_0 takes 735 GB, and unquantized FP16 weights take 1370 GB.
Can DeepSeek V3.2 run without a GPU?
Yes — with roughly 512 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.
What context window does DeepSeek V3.2 support?
DeepSeek V3.2 supports a 125k-token context window (128,000 tokens).
Can I use DeepSeek V3.2 commercially?
Yes. DeepSeek V3.2 is released under MIT, a permissive open-source license that allows commercial use, modification and redistribution.
How fast is DeepSeek V3.2 on consumer hardware?
Our compatibility engine estimates on the order of 5 tokens/sec on a mid-range GPU and up to 15 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.
Which quantization of DeepSeek V3.2 should I download first?
Start with Q4_K_M (410 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It does not fit a single 24 GB consumer card — plan for multi-GPU or server hardware.