DeepSeek-R1-0528 (671B)
DeepSeek-R1-0528 (671B) by DeepSeek (671B parameters, MoE with 37B active per token). At Q4_K_M with an 8K context it needs about 398.8 GiB of memory. The full 671B DeepSeek-R1 (37B active per token): even at 4-bit it needs more memory than any single machine in our catalog.
Last updated: 2026-10-13
Model facts
Run it
ollama run deepseek-r1:671b
Memory needed
Weights + KV cache + runtime overhead, in GiB. Lower quantization = less memory, slightly lower quality.
| Quantization | 4K context | 8K context | 32K context |
|---|---|---|---|
| Q4_K_M | 398.6 GiB | 398.8 GiB | 400.4 GiB |
| Q5_K_M | 467.5 GiB | 467.7 GiB | 469.3 GiB |
| Q6_K | 541.3 GiB | 541.6 GiB | 543.2 GiB |
| Q8_0 | 697.9 GiB | 698.2 GiB | 699.8 GiB |
| FP16 | 1313.1 GiB | 1313.4 GiB | 1315.0 GiB |
Which hardware runs it
Best quantization that fits each device with a 8K context. Estimates, not benchmarks. See the methodology.
Frequently asked questions
How much VRAM does DeepSeek-R1-0528 (671B) need?
With an 8K context, about 398.8 GiB at Q4_K_M and 698.2 GiB at Q8_0 (weights + KV cache + runtime overhead). Longer contexts need more memory for the KV cache.
Can DeepSeek-R1-0528 (671B) run on an GeForce RTX 4090?
DeepSeek-R1-0528 (671B) does not fit on the GeForce RTX 4090 (24 GB), even at Q4_K_M with an 8K context.
Can I run DeepSeek-R1-0528 (671B) on a Mac?
DeepSeek-R1-0528 (671B) does not fit on the M3 Ultra (512 GB), even at Q4_K_M with an 8K context.
Can DeepSeek-R1-0528 (671B) be used commercially?
DeepSeek-R1-0528 (671B) is released under MIT. The licence allows commercial use (read its conditions).
Sources
- https://api-docs.deepseek.com/news/news250528
- https://huggingface.co/deepseek-ai/DeepSeek-R1-0528
- https://huggingface.co/deepseek-ai/DeepSeek-R1
- https://huggingface.co/deepseek-ai/DeepSeek-R1-0528/resolve/main/config.json
- https://huggingface.co/api/models/deepseek-ai/DeepSeek-R1-0528?expand[]=safetensors
- https://ollama.com/library/deepseek-r1/tags