Microsoft4 local models

Phi Models: Small AI That Thinks Big

Microsoft's Phi family proves that size is not everything. Phi-4 Mini at just 3.8B parameters matches or beats many 7B models on reasoning benchmarks. If you have a device with limited RAM (8-16GB), Phi models give you the most intelligence per gigabyte.

Developer

Microsoft

Models

4

Size Range

3.8B – 14B

RAM Range

7 – 22 GB

Key Features

✓Best quality-per-parameter in small sizes
✓Strong reasoning for its size
✓Phi-4 Mini rivals 7B models at 3.8B
✓Runs well on low-RAM devices

All Phi Models

ModelSizeQuantVRAMMin RAMBest ForQualityOllama
Phi-3 Mini 3.8B3.8BQ4_K_M3.2 GB7 GBCoding, Chat
70
Phi-4 Mini 3.8B3.8BQ4_K_M3.2 GB7 GBCoding, Chat
72
Phi-3 Medium 14B14BQ4_K_M11 GB20 GBCoding, Quality
89
Phi-4 14B14BQ4_K_M11.5 GB22 GBCoding, Quality
92

Device Compatibility

Which Phi models can run on each device class, based on minimum RAM requirements.

ModeliPhoneAirProStudioMini
Phi-3 Mini 3.8B (3.8B)PossiblePossibleExcellentExcellentExcellent
Phi-4 Mini 3.8B (3.8B)PossiblePossibleExcellentExcellentExcellent
Phi-3 Medium 14B (14B)NoPossiblePossibleExcellentPossible
Phi-4 14B (14B)NoPossiblePossibleExcellentPossible

RAM Requirements

Phi-3 Mini 3.8B
3.2 GB
min 7 GB
Phi-4 Mini 3.8B
3.2 GB
min 7 GB
Phi-3 Medium 14B
11 GB
min 20 GB
Phi-4 14B
11.5 GB
min 22 GB

Frequently Asked Questions

Is Phi-4 Mini really as good as 7B models?+
On reasoning and math benchmarks, yes. Phi-4 Mini 3.8B matches Llama 3.2 8B on several tasks while using half the RAM. For pure chat quality, 7B models still have a slight edge.
What RAM does Phi need?+
Phi-4 Mini Q4 needs just 7GB RAM. Phi-4 14B needs 22GB. The Mini model is perfect for 8GB MacBook Airs and iPhones.
How does Phi compare to Gemma?+
Phi-4 Mini (3.8B) vs Gemma 2 2B: Phi wins clearly. Phi-4 14B vs Gemma 2 27B: Gemma wins on quality but needs more RAM. Pick Phi when RAM is tight.

Related Model Families

GemmaGoogle DeepMindQwenAlibaba CloudLlamaMeta

Getting Started

How to Set Up OllamaBest LLM for iPhoneBrowse All DevicesModelFit Wizard