google/gemma-3-4b

gemma3 · 4B · 4bit

MacBook Air (Apple M4)

32 GB · macOS 26.3

Tested on March 3, 2026

Top 25% Compare

Global Score

74 /100

Good

Hardware Fit

93/100

Quality

66/100

Get this model

🦙

Ollama

ollama pull google/gemma-3-4b

View on Ollama Library

ollama.com/library/google/gemma-3-4b

🤗

Find on HuggingFace

GGUF versions & quantizations

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: 22.2
Standard deviation Variation of speed across benchmark runs — lower means more consistent: ±15.4
Time to first token How fast the model starts responding — lower is better: 1.9 s
Load time Time to load the model into memory before inference: 0.0 s
Memory usage RAM consumed during inference — percentage of total system memory: 0.0 GB (0%)
Total tokens Total number of tokens generated across all benchmark prompts: 808

Score breakdown

Speed

35/40

Time to first token

28/30

Memory

30/30

Quality

Reasoning

10/20

Coding

17/20

Instruction following

13/20

Structured output

12/15

Math

4/15

Multilingual

10/10

Category levels

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

Metadata

Spec version: 0.2.0
Runtime: Lm-studio unknown
Model format: GGUF
Hardware profile: BALANCED
Result hash: 9b270d6a9dc03fca64ec1be8348eede9536ed7103944f11449fe30dfe00df3ef

Interpretation

Hardware fit: 93/100. Overall suitability: GOOD (Global 74/100). Category profile: Reasoning: Adequate, Coding: Strong, Instruction Following: Adequate, Structured Output: Strong, Math: Weak, Multilingual: Strong.

Warnings

Token speed is unstable (stddev 15.4 tok/s, mean 22.2 tok/s) — may indicate thermal throttling or memory pressure.

Run yours now

npx metrillm@latest bench

Requires Node 20+ and Ollama or LM Studio running