phi-3-mini-128k-instruct

phi3 · 4bit

MacBook Air (Apple M4)

32 GB · macOS 26.3

Tested on March 4, 2026

Top 93% Compare

Global Score

38 /100

Not Rec.

Hardware Fit

100/100

Quality

12/100

Get this model

Get it in LM Studio

Search and download models directly from the app

🤗

Find on HuggingFace

MLX versions & conversions

Hardware

Machine: MacBook Air
CPU: Apple M4
Cores: 10 total (4 perf + 6 eff)
Frequency: 2.4 GHz
RAM: 32 GB LPDDR5
GPU: Apple M4
OS: macOS 26.3
Arch: arm64
Power Mode: balanced

Performance

Tokens/sec Tokens generated per second — higher is better: 44.0
Standard deviation Variation of speed across benchmark runs — lower means more consistent: ±41.7
First chunk latency Delay before the first streamed chunk arrives from runtime: 384 ms
Time to first token How fast the model starts responding — lower is better: 385 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 (6%)
Total tokens Total number of tokens generated across all benchmark prompts: 1014

Score breakdown

Speed

50/50

Time to first token

20/20

Memory

30/30

Quality

Reasoning

3/20

Coding

2/20

Instruction following

2/20

Structured output

1/15

Math

2/15

Multilingual

2/10

Category levels

Reasoning: Poor Coding: Poor Instruction Following: Poor Structured Output: Poor Math: Poor Multilingual: Poor

Metadata

Spec version: 0.2.1
Runtime: LM Studio 0.4.6+1
Model format: MLX
Hardware profile: BALANCED
Result hash: 1974f2fead57bf5f257a3f233fa160e4ec3879298f9b1058cd7088a640a0d79a

Interpretation

Hardware fit: 100/100. Overall suitability: NOT RECOMMENDED (Global 38/100). Category profile: Reasoning: Poor, Coding: Poor, Instruction Following: Poor, Structured Output: Poor, Math: Poor, Multilingual: Poor. Warning: model produced very low accuracy on quality tasks — results may be unusable despite good hardware performance.

Warnings

Model produced very low accuracy on quality tasks — results may be unusable despite good hardware performance.
Token speed is unstable (stddev 41.7 tok/s, mean 44.0 tok/s) — may indicate thermal throttling or memory pressure.

Bench Environment

Power: AC CPU load: avg 17% (peak 25%)

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