DeepSeek · open-weights model
DeepSeek-R1
DeepSeek-R1 is a 671B-parameter model (MoE, 37B active) by DeepSeek. Its smallest quant (Q4_K_M) needs at least 405.6GB of GPU-addressable memory at 8K context; none of the 50 machines we track load it cleanly — that is the honest answer.
- Parameters
- 671B total · 37B active
- Layers
- 61
- Max context
- 160K
- Attention
- MLA
- License
- mit
Config from deepseek-ai/DeepSeek-R1
Quants and what they need
Measured file sizes; “needs” = weights + KV cache at 8K context + overhead.
| Quant | File size | Needs ≥ |
|---|---|---|
| Q8_0 | 713.3 GB | 714.5 GB |
| Q6_K | 550.8 GB | 552.0 GB |
| Q5_K_M | 475.4 GB | 476.6 GB |
| Q4_K_M | 404.4 GB | 405.6 GB |
What runs DeepSeek-R1
Every machine we track, graded at its best clean-fit quant.
Verdicts computed from published specs and measured GGUF sizes — how the math works.