DeepSeek V4 family: which size to run
2 sizes, 284B–1.6T parameters · DeepSeek
Sizes
| Model | Architecture | |||||||
|---|---|---|---|---|---|---|---|---|
| DeepSeek V4 Flash | 284B | MoE · 13B active | 1,048,576 | 65.5 | 221 | 215 | No GGUF Q4_K_M file (plain or UD-) in the repository collected | 160.1 |
| DeepSeek V4 Pro | 1.6T | MoE · 49B active | 1,048,576 | 71.6 | 1,208 | 1,202 | No file facts collected for this model | No file facts collected for this model |
Each size is its own model, with its own scores and memory needs. Q4 estimates are planning figures, not measurements; file sizes are published. The Mac or PC column is for a 32 GB or larger machine (8 GB reserve); machines under 32 GB reserve 25% of their memory (2 to 8 GB), so they need less. The publisher lists V4-Pro 1.6T (49B active), V4-Flash 284B (13B active).
The two sizes listed are the preview checkpoints. DeepSeek later published official releases of both (below), each with its own scores. All are very large mixture-of-experts models.
How to read this table
Q4 means compressed to about 4 bits per weight. GGUF (llama.cpp, Ollama, LM Studio) and MLX (Apple silicon) are file formats; the sizes shown are the published 4-bit files. A dash means no value is published or collected for that size; the reason is in the cell's accessible text.
Which size fits my machine
The largest size whose Q4 planning estimate fits what each machine offers. A planning estimate, not a measurement and not a speed. Smaller machines are left out; the largest one that fits no size is kept to show the limit.
| Machine | Largest DeepSeek V4 size with enough memory | ||
|---|---|---|---|
| Mac Studio M5 Max · 128 GB unified | 104 | No size of this family has a Q4 estimate within what this machine offers | No size fits, so there is no estimate to compare |
| Mac Studio · M5 Ultra · 36-core CPU / 80-core GPU · 512 GB unified | 392 | DeepSeek V4 Flash | 221 |
| RTX 5090 · 32 GB VRAM | 32 | No size of this family has a Q4 estimate within what this machine offers | No size fits, so there is no estimate to compare |
| DGX H100 · 8 × 80 GB | 640 | DeepSeek V4 Flash | 229 |
A dash means no size of this family fits that machine. The requirement is the size's Q4 estimate with that machine's own reserve: a graphics card reserves 2 GB, as in the sizes table; a Mac or PC reserves 25% of its memory (2 to 8 GB), so below 32 GB its figure is lower than the sizes table's. A server profile reserves 2 GB per graphics card, so its figure is higher.
Smallest machine for each size
The profile with the least memory that still has enough for the size's Q4 estimate, from the hardware catalog.
| Size | Smallest Mac or PC-memory profile | Smallest graphics-card profile | Server-class profile |
|---|---|---|---|
| DeepSeek V4 Flash | Mac Studio 2026 · M5 Ultra · 512 GB | No graphics-card profile in the catalog has enough memory for this size | DGX H100 · 8 × 80 GB · 640 GB |
| DeepSeek V4 Pro | No Mac, laptop or PC profile in the catalog has enough memory for this size | No graphics-card profile in the catalog has enough memory for this size | Even the largest profile in the catalog (DGX H100 · 8 × 80 GB, 640 GB aggregated) is below this planning estimate |
A dash means no profile of that kind in the catalog has enough memory for the size.
Memory by file precision
The same planning formula at three precisions (0.625, 1.125 and 2 bytes per parameter, plus working room and the system reserve). Estimates, not measurements.
| Size | ||||||
|---|---|---|---|---|---|---|
| DeepSeek V4 Flash | 221 | 215 | 392 | 386 | 690 | 684 |
| DeepSeek V4 Pro | 1,208 | 1,202 | 2,168 | 2,162 | 3,848 | 3,842 |
Family tree
Family, sizes and, for each size, the two 4-bit files the tables use (GGUF Q4_K_M and MLX 4-bit) when one was collected. Other precisions are not drawn here.
The same tree as a list:
- DeepSeek V4 · DeepSeek
- Flash 284B · 13B active
- MLX 4-bit · 160.1 GB · mlx-community/deepseek-ai-DeepSeek-V4-Flash-4bit, read Oct 23, 2026
- Pro 1.6T · 49B active
- No files collected
- Flash 284B · 13B active
Learn more
See DeepSeek V4 Flash
The smallest size of the family: its scores, memory estimate and the machines it runs on.
Sources
Where the family and its sizes are named, and where each number comes from.
- deepseek-ai/DeepSeek-V4-Flash model card · “We present a preview version of DeepSeek-V4 series, including two strong Mixture-of-Experts (MoE) language models — DeepSeek-V4-Pro with 1.6T parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated) — both supporting a context length of one million tokens.” · read Oct 17, 2026
- deepseek-ai/DeepSeek-V4-Flash-0731 model card · “DeepSeek-V4-Flash-0731 is the official release of DeepSeek-V4-Flash, superseding the preview version” · read Oct 17, 2026
- deepseek-ai/DeepSeek-V4-Pro-0813 model card · “DeepSeek-V4-Pro-0813 is the official release of DeepSeek-V4-Pro, superseding the preview version” · read Oct 17, 2026
- deepseek-ai/DeepSeek-V4-Flash-Vision-Exp model card · “It builds on the DeepSeek-V4-Flash architecture by incorporating visual modules and undergoing continued training to unlock visual understanding capabilities.” · read Oct 17, 2026
- deepseek-ai/DeepSeek-V4.1-Flash model card · “DeepSeek-V4.1-Flash is trained from scratch on a multimodal corpus comprising 45T tokens” · read Oct 17, 2026
- DeepSeek V4 Flash model card
- DeepSeek V4 Pro model card
- LiveBench 2026-06-25 · the only quality source on this page; a dash means the release does not include the model · snapshot collected Sep 26, 2026
- Published file sizes: Hugging Face repositories, read Oct 23, 2026
- Methodology · how the Q4 planning estimate is computed · Learn