Phi-4 family: which size to run
2 sizes, 3.8B–14B parameters · Microsoft
Sizes
| Model | Architecture | |||||||
|---|---|---|---|---|---|---|---|---|
| Phi-4 mini | 3.8B | Dense | 131,072 | No LiveBench result for this model in release 2026-06-25 | 11 | 5 | 2.5 | 2.2 |
| Phi-4 | 14B | Dense | 16,384 | No LiveBench result for this model in release 2026-06-25 | 19 | 13 | The repository labels its file Q4_K (llama.cpp's alias for Q4_K_M); the site's file ladder matches only the Q4_K_M label and its UD- variant | 8.2 |
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 two cards describe models with different designs: the mini card says it “employed new architecture for efficiency, larger vocabulary for multilingual support”. Compare them as two models, not as one model at two sizes.
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.
| Machine | Largest Phi-4 size with enough memory | ||
|---|---|---|---|
| Mac mini M2 · 8 GB unified | 8 | Phi-4 mini | 5 |
| Mac mini M4 · 16 GB unified | 16 | Phi-4 | 15 |
| Mac mini M4 · 24 GB unified | 24 | Phi-4 | 17 |
| Mac mini M4 · 32 GB unified | 32 | Phi-4 | 19 |
| Mac mini M5 Pro · 64 GB unified | 56 | Phi-4 | 19 |
| Mac Studio M5 Max · 128 GB unified | 104 | Phi-4 | 19 |
| Mac Studio · M5 Ultra · 36-core CPU / 80-core GPU · 512 GB unified | 392 | Phi-4 | 19 |
| RTX 4060 · 8 GB VRAM | 8 | Phi-4 mini | 5 |
| RTX 3060 · 12 GB VRAM | 12 | Phi-4 mini | 5 |
| RTX 5080 · 16 GB VRAM | 16 | Phi-4 | 13 |
| RTX 4090 · 24 GB VRAM | 24 | Phi-4 | 13 |
| RTX 5090 · 32 GB VRAM | 32 | Phi-4 | 13 |
| DGX H100 · 8 × 80 GB | 640 | Phi-4 | 27 |
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 |
|---|---|---|
| Phi-4 mini | Mac mini 2020 · M1 · 8 GB | RTX 3050 · 6 GB |
| Phi-4 | Mac mini 2020 · M1 · 16 GB | RTX 4060 Ti · 16 GB |
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 | ||||||
|---|---|---|---|---|---|---|
| Phi-4 mini | 11 | 5 | 14 | 8 | 18 | 12 |
| Phi-4 | 19 | 13 | 27 | 21 | 42 | 36 |
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:
- Phi-4 · Microsoft
- mini 3.8B
- GGUF Q4_K_M · 2.5 GB · unsloth/Phi-4-mini-instruct-GGUF, read Oct 23, 2026
- MLX 4-bit · 2.2 GB · mlx-community/Phi-4-mini-instruct-4bit, read Oct 23, 2026
- 14B
- MLX 4-bit · 8.2 GB · mlx-community/phi-4-4bit, read Oct 23, 2026
- mini 3.8B
Learn more
See Phi-4 mini
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.
- microsoft/Phi-4-mini-instruct model card · “The model belongs to the Phi-4 model family and supports 128K token context length.” · read Oct 17, 2026
- microsoft/Phi-4-mini-instruct model card · “The Phi-4-mini model employed new architecture for efficiency, larger vocabulary for multilingual support, and better post-training techniques were used for instruction following” · read Oct 17, 2026
- microsoft/phi-4 model card · “Architecture: 14B parameters, dense decoder-only Transformer model” · read Oct 17, 2026
- Phi-4 mini model card
- Phi-4 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