microsoft/phi-4-mini-reasoning

phi3 · 3.8B · 4bit

THINKING MODEL

Mac mini (Apple M4 Pro)

64 GB · macOS 15.7.4

Tested on March 5, 2026

Top 58% Compare

Global Score

64 /100

Good

Hardware Fit

100/100

Quality

48/100

Get this model

Get it in LM Studio

Search and download models directly from the app

🤗

Find on HuggingFace

MLX versions & conversions

Hardware

Machine: Mac mini
CPU: Apple M4 Pro
Cores: 14 total (10 perf + 4 eff)
Frequency: 2.4 GHz
RAM: 64 GB LPDDR5
GPU: Apple M4 Pro
OS: macOS 15.7.4
Arch: arm64
Power Mode: balanced

Performance

Tokens/sec Tokens generated per second — higher is better: 90.3
Standard deviation Variation of speed across benchmark runs — lower means more consistent: ±0.1
First chunk latency Delay before the first streamed chunk arrives from runtime: 198 ms
Time to first token How fast the model starts responding — lower is better: 198 ms
Load time Time to load the model into memory before inference: N/A
Memory usage RAM consumed during inference — percentage of total system memory: 2.0 GB (3%)
Total tokens Total number of tokens generated across all benchmark prompts: 1463

Score breakdown

Speed

50/50

Time to first token

20/20

Memory

30/30

Quality

Reasoning

7/20

Coding

7/20

Instruction following

7/20

Structured output

7/15

Math

13/15

Multilingual

7/10

Category levels

Reasoning: Weak Coding: Weak Instruction Following: Weak Structured Output: Weak Math: Strong Multilingual: Adequate

Metadata

Spec version: 0.2.1
Runtime: LM Studio 0.4.6+1
Model format: MLX
Hardware profile: HIGH-END
Result hash: f3d0a02823295c7a48c1df8305e2ef8a3fdd7b58b95a83d1b35f3ece121c4c66

Interpretation

Hardware fit: 100/100. Overall suitability: GOOD (Global 64/100). Category profile: Reasoning: Weak, Coding: Weak, Instruction Following: Weak, Structured Output: Weak, Math: Strong, Multilingual: Adequate.

Bench Environment

Power: AC CPU load: avg 24% (peak 26%)

Run yours now

$ npm install -g metrillm@latest

$ metrillm

Requires Node 20+ and Ollama or LM Studio running

Or run without installing: npx metrillm@latest