Can I run DeepSeek V3.2 locally?
DeepSeek V3.2 by deepseek-ai has 685B parameters (41B active per token, 256 experts, 8 used per token). At 4-bit with an 8K context it needs about 379 GB of GPU memory. It fits on 1 of 65 GPUs and Macs we track.
Memory needed by quant and context length
GB of GPU memory = weights + KV cache (FP16) + runtime overhead. Weights use real file sizes from unsloth/DeepSeek-V3.2-GGUF where available (marked •), otherwise parameter count × bits per weight. How we calculate.
| Quant | Quality | 4K ctx | 8K ctx | 32K ctx | 128K ctx |
|---|---|---|---|---|---|
| 2-bit (Q2_K) • | rough | 229 GB | 230 GB | 232 GB | 240 GB |
| 3-bit (Q3_K_M) • | usable | 299 GB | 300 GB | 302 GB | 309 GB |
| 4-bit (Q4_K_M) • | good, most popular | 379 GB | 379 GB | 381 GB | 389 GB |
| 5-bit (Q5_K_M) • | very good | 444 GB | 445 GB | 447 GB | 455 GB |
| 6-bit (Q6_K) • | near lossless | 514 GB | 515 GB | 517 GB | 525 GB |
| 8-bit (Q8_0) • | effectively lossless | 665 GB | 666 GB | 668 GB | 675 GB |
| 16-bit (F16/BF16) • | original weights | 1251 GB | 1251 GB | 1253 GB | 1261 GB |
Which GPUs and Macs run DeepSeek V3.2?
Best quality quant that fits at 8K context. "Tight" = within 10% of usable memory. Speed is a ceiling from memory bandwidth (real speed is typically 50–80% of it).
Prefer an API?
DeepSeek: DeepSeek V3.2 is on OpenRouter at $0.26 per 1M input tokens and $0.38 per 1M output tokens. Price details and history →
Questions
How much VRAM does DeepSeek V3.2 need?
About 379 GB at 4-bit (Q4_K_M) with an 8K context: 378 GB of weights (real GGUF file size), 0.7 GB of KV cache and 0.6 GB of runtime overhead. At 8-bit it needs about 666 GB.
What is the smallest GPU that runs DeepSeek V3.2?
No single consumer GPU in our table fits it at Q4_K_M with 8K context.
Can I run DeepSeek V3.2 on a Mac?
Not at Q4_K_M with 8K context on any Mac in our table, even a 512 GB Mac Studio, using the default GPU memory limit.
Model facts
| Hugging Face repo | deepseek-ai/DeepSeek-V3.2 |
| Parameters | 685,355,329,792 (685B) · ~41B active (computed from config) |
| Layers / hidden size | 61 / 7168 |
| Attention | MLA; 128 heads, 128 KV heads, head dim 56 |
| KV cache per token (FP16) | 83.9 KiB |
| Max context | 163,840 tokens |
| License | mit |
| Released on HF | Dec 1, 2025 |
| Downloads (30d) / likes | 9,203,313 / 1,602 |
Sources: Hugging Face model API · config.json (deepseek-ai/DeepSeek-V3.2) · GGUF files · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.