winnow-12b
LM-STUDIO GGUFgemma4 · 12B · Q8_0
Nov 1, 2026 · AMD RYZEN AI MAX+ 395
qwen3.5-4b-mlx
LM-STUDIO MLXqwen3_5 · 4B · 4bit
Mar 3, 2026 · Apple M4
Global Score
77 vs 71
Hardware Fit
38 vs 88
Quality Score
93 vs 63
Hardware
winnow-12b qwen3.5-4b-mlx
MachineGMKtec EVO-X3MacBook Air
CPUAMD RYZEN AI MAX+ 395Apple M4
Cores3210
RAM120 GB32 GB
GPUAMD Radeon(TM) 8060S GraphicsApple M4
OSMicrosoft Windows 11 Pro 10.0.26300macOS 26.3
Archx64arm64
Power Modebalancedbalanced
Performance
winnow-12b qwen3.5-4b-mlx
Tokens/sec12.923.1
First chunk19 ms3279 ms
TTFT9.7 s3.3 s
Load time86.2 sN/A
Memory usage51.3 GB2.9 GB
Memory %43%9%
HW Fit Score Breakdown
winnow-12b
Speed
11/50
TTFT
0/20
Memory
27/30
qwen3.5-4b-mlx
Speed
45/50
TTFT
13/20
Memory
30/30
Quality
winnow-12b
Reasoning
19/20
Coding
19/20
Instruction
18/20
Structured
15/15
Math
12/15
Multilingual
10/10
Reasoning: Strong Coding: Strong Instruction Following: Strong Structured Output: Strong Math: Strong Multilingual: Strong
qwen3.5-4b-mlx
Reasoning
15/20
Coding
16/20
Instruction
6/20
Structured
6/15
Math
11/15
Multilingual
9/10
Reasoning: Adequate Coding: Strong Instruction Following: Weak Structured Output: Weak Math: Adequate Multilingual: Strong
Run yours and compare
$
npm install -g metrillm@latest$
metrillmRequires Node 20+ and Ollama or LM Studio running
Or run without installing: npx metrillm@latest