DeepSeek V4 Models — Hardware Requirements
13 DeepSeek V4 models from DeepSeek and the community, from the smallest that runs in 17.9 GB of VRAM at Q4 up to 1650.5B parameters. Every row links to full quantization tables and GPU compatibility.
All DeepSeek V4 Models by Size
| Model | Params | Runs from (Q4) | Context | Publisher | Quant downloads |
|---|---|---|---|---|---|
| DeepSeek V4 Flash | 290.9B | 175.1 GB | 1049K | ||
| DeepSeek V4 Flash 0731 | 304.2B | 183.0 GB | 1049K | ||
| DeepSeek V4 Flash DSpark | 304.2B | 183.0 GB | 1049K | ||
| DeepSeek V4 Flash Vision Exp | 304.6B | 183.3 GB | 1049K | ||
| DeepSeek V4.1 Flash | 763.2B | 458.4 GB | 1049K | ||
| DeepSeek V4.1 Flash 1017 | 763.2B | 458.4 GB | 1049K | ||
| DeepSeek V4 Pro | 1598.8B | 959.8 GB | 1049K | ||
| DeepSeek V4 Pro 0813 | 1650.5B | 990.8 GB | 1049K | ||
| DeepSeek V4 Pro DSpark | 1650.5B | 990.8 GB | 1049K |
Community variants and fine-tunes(14)
| Model | Params | Runs from (Q4) | Context | Publisher | Quant downloads |
|---|---|---|---|---|---|
| EVE 27B XENO HAT DeepSeek V4 Flash | 27.8B | 17.9 GB | 262K | ||
| DeepSeek V4 Flash JANG CRACK | 33.5B | 20.6 GB | 1049K | ||
| Deepseek V4 Flash 0731 Spark | 60.3B | 36.7 GB | 1049K | ||
| DeepSeek V4 Flash 162B | 92.2B | 55.8 GB | 1049K | ||
| DeepSeek V4 Flash 180B | 101.6B | 61.4 GB | 1049K | ||
| DeepSeek V4.1 Flash 1017 EXL3 2.9bpw | 105.3B | 63.7 GB | 1049K | ||
| DeepSeek V4 Flash 0731 Reap 150B | 150.1B | 90.6 GB | 1049K | ||
| DeepSeek V4 Flash DSpark Abliterated Uncensored | 165.3B | 115.1 GB | — | ||
| Keys DeepSeekV4 Flash GA 0731 Dspark Abliterated Anchored Tensors | 304.2B | 207.9 GB | — | ||
| DeepSeek V4 Flash Vision Uncensored | 304.6B | 208.2 GB | — | ||
| DeepSeek V4 Flash 0731 Abliterated | 305.7B | 183.9 GB | 1049K | ||
| SuperDeepseek V4 Flash Abliterated MQ 2xDGX | 306.3B | 184.2 GB | 1049K | ||
| DeepSeek V4.1 Flash 1017 Obliterated | 754.6B | 453.3 GB | 1049K | ||
| DeepSeek V4.1 Flash 1017 MXFP4 Uncensored | 763.2B | 458.4 GB | 1049K |
How DeepSeek V4 Compares — llmrun Composite Score
DeepSeek V4.1 Flash 1017 is the highest-scoring DeepSeek V4 model with an llmrun Composite Score of 134.4 — #2 among 106 open models. The top proprietary model, Gemini 4 Carbon, scores 158.6. Click a model to see its full benchmark breakdown.
Gemini 4 Carbon · proprietary158.6
GPT 6 Astra · proprietary156.2
Claude Opus 5.5 · proprietary153.0
GPT 6.1 Sol · proprietary151.7
GPT 6.1 Terra · proprietary150.1
Kimi K3.1 Pro142.8
MiMo V2.6 Pro RL133.3
Kimi K3132.7
MiMo V2.6 Flash RL131.9
GLM 5.3130.1
DeepSeek V4.1 Flash129.4
DeepSeek V4 Pro 0813129.2
Step 5 Preview128.0
GLM 5.3 Flash124.1
DeepSeek V4 Pro116.7
DeepSeek V4 Flash115.3
Frequently Asked Questions
- How much VRAM do I need to run a DeepSeek V4 model?
- The smallest DeepSeek V4 model, EVE 27B XENO HAT DeepSeek V4 Flash, runs from 17.9 GB of VRAM at Q4_K_M (smaller low-bit quantizations start at 12.6 GB with a noticeable quality loss). Larger family members need proportionally more — see the table above for every model.
- Which DeepSeek V4 models can I run on a 16 GB GPU?
- No DeepSeek V4 model currently fits in 16 GB of VRAM — the family starts at 17.9 GB.
- What is the most popular DeepSeek V4 model to run locally?
- DeepSeek V4.1 Flash is the most downloaded DeepSeek V4 model in local-friendly quantized formats. It runs from 458.4 GB of VRAM at Q4_K_M.
- How do DeepSeek V4 models score on benchmarks?
- DeepSeek V4.1 Flash 1017 leads the family with an llmrun Composite Score of 134.4, ranking #2 among 106 open models, while the top proprietary model, Gemini 4 Carbon, scores 158.6. See the comparison chart above for the full standings.