Skip to content
localmodel.run

Text model · phi

Phi-4-mini 3.8B requirements

phi family · 3.8B params · released Feb 2025 · 931.9K Ollama pulls. Minimum to run at Q4_K_M: Apple M1 (8GB).

LicenseMIT· Commercial OK↓ 948.2K/mo♥ 766on HuggingFace
Q4_K_M
2.49 GB
Q8_0
4.08 GB
Total @ Q4 (4k)
~3.8 GB
Context
128 k

Quantization sizes

GGUF quantson disk
QuantizationSize on disk
Q2_K1.6 GB est
Q3_K_M1.9 GB est
Q4_K_M (default)2.49 GB
Q5_K_M2.7 GB est
Q6_K3.1 GB est
Q8_04.08 GB
FP167.6 GB est

Lower quant = smaller and faster, slightly lower quality. Q4_K_M is the common default.

Run it

Ollama
$ ollama run phi4-mini:3.8b
llama.cpp
$ llama-cli -hf bartowski/microsoft_Phi-4-mini-instruct-GGUF:Q4_K_M
LM Studio
$ lms get bartowski/microsoft_Phi-4-mini-instruct-GGUF

Which devices can run Phi-4-mini 3.8B?

FAQ

How much VRAM or RAM does Phi-4-mini 3.8B need?

At Q4_K_M, Phi-4-mini 3.8B needs about 3.8 GB (weights ~2.49 GB + KV cache + overhead) at a 4k context. At Q8_0 budget ~5.4 GB.

Can Phi-4-mini 3.8B run on a laptop?

Yes, Phi-4-mini 3.8B fits on a 16 GB machine at Q4_K_M and runs on Apple Silicon or a 12 GB+ GPU comfortably.

Can I use Phi-4-mini 3.8B commercially?

Yes. Phi-4-mini 3.8B is licensed MIT, which permits commercial use.

Released February 2025 by Microsoft. Dense decoder-only transformer with 200K vocabulary. 128K context. Q4_K_M=2.49GB and Q8_0=4.08GB from bartowski HF repo (microsoft_Phi-4-mini-instruct-GGUF). Requires Ollama 0.5.13+.

Sources

Memory figures are estimates. See methodology.