Model record
Phi-4 Mini
Microsoft · MIT · verified
What this model is
The architecture is dense. Every parameter is part of the active path, so memory and speed scale with the full model rather than a smaller expert subset.
Memory and quantization
Decimal GB, rounded from the Ollama tag page. Runtime memory still needs KV cache and overhead.
| Quantization | Published files | 8K estimate | Smallest comfortable GPU | Smallest comfortable Mac | Context at smallest GPU / Mac | Speed estimate | Run commands |
|---|---|---|---|---|---|---|---|
| Q4_K_M | 2.50 GB | 5.07 GB | NVIDIA GeForce RTX 3060 12GB · 12 GB VRAM | Apple Mac mini M4 16GB · 16 GB unified memory | 61,037 / 91,556 tokens | Estimate · Fast relative class | ollama run phi4-mini:3.8b-q4_K_M lms get "Phi-4 Mini"@q4_k_m |
| Q8_0 | 4.10 GB | 6.67 GB | NVIDIA GeForce RTX 3060 12GB · 12 GB VRAM | Apple Mac mini M4 16GB · 16 GB unified memory | 48,830 / 79,348 tokens | Estimate · Fast relative class | ollama run phi4-mini:3.8b-q8_0 lms get "Phi-4 Mini"@q8_0 |
Estimate formula: published weight files + 1.0737 GB per 8K tokens × context / 8K + 1.5 GB runtime overhead. Sizes are decimal GB, not GiB. At 8K context, the smallest listed build is about 5.1 GB before OS and application reserve.
Conservative FP16 estimate from the Phi-4 Mini GQA configuration; not a runtime measurement.
Ollama tags were checked against the linked Ollama library pages on 2026-11-02. LM Studio's lms get suffix selects a file when the search results expose that exact quantization; confirm the local key with lms ls.
KV method source: https://ollama.com/library/phi4-mini
Speed heuristic
Speed depends on quantization, backend, and hardware; no single measured value is reported
Active path: 3.8B parameters. This is a relative speed signal only; it is not a tokens/sec benchmark.
Runtimes
- Ollama
- llama.cpp
ollama run phi4-minilms get "Phi-4 Mini"@q4_k_m
lms load <model-key-from-lms-ls> --gpu max --context-length 8192License restrictions
MIT permits commercial use and modification; preserve the copyright and permission notice.
This is a plain-language summary, not legal advice. Read the linked license before commercial deployment or redistribution.
Known issues
- Not intended for long-context or highly complex tasks
If it does not fit, try these smaller records
These alternatives have a lower recommended memory target in the current catalog. Open each page to compare the quantization-specific fit before downloading.
Llama 3.2 3B
Dense · starting at 4 GB
FAQ
Model-specific questions
What is the smallest local file for Phi-4 Mini?
The smallest published weight package in this record is Q4_K_M at 2.50 GB. The 8K runtime estimate adds 1.07 GB of KV cache and 1.5 GB overhead, for about 5.1 GB before operating-system reserve.
How much memory do the published quantizations need at 8K?
Q4_K_M about 5.1 GB at 8K; Q8_0 about 6.7 GB at 8K. These are calculated estimates using the published weight sizes and the model's KV-cache assumption, not live memory measurements.
Which GPU and Mac configurations are comfortable for Phi-4 Mini?
Against the current catalog, the smallest configurations that clear the 20% headroom target are: Q4_K_M: NVIDIA GeForce RTX 3060 12GB; Q8_0: NVIDIA GeForce RTX 3060 12GB. A larger configuration may still be preferable for longer context or faster memory bandwidth.
What is the right Ollama or LM Studio command for Phi-4 Mini?
Use the command attached to the specific quantization in the table. Ollama tags are taken from the checked Ollama library pages. For LM Studio, use lms get with the displayed quantization suffix, then load the resulting local key from lms ls with --gpu max and the desired context length.
What should I run instead if Phi-4 Mini does not fit?
Smaller catalog alternatives include Llama 3.2 3B. Compare their quantization tables and minimum memory before choosing; a smaller parameter count does not always mean a proportionally smaller download.
Sources
This page is an independent compatibility guide, not official documentation. Verify the primary source before downloading or deploying a model.