qwen3.5-0.8b-mlx
LM-STUDIO MLXqwen3_5 · 0.8B · 4bit
Mar 3, 2026 · Apple M4
qwen2.5-math-1.5b-instruct
LM-STUDIO MLXqwen2 · 1.5B · 4bit
Mar 4, 2026 · Apple M4
Global Score
55 vs 45
Hardware Fit
87 vs 99
Quality Score
41 vs 22
Hardware
qwen3.5-0.8b-mlx qwen2.5-math-1.5b-…
MachineMacBook AirMacBook Air
CPUApple M4Apple M4
Cores1010
RAM32 GB32 GB
GPUApple M4Apple M4
OSmacOS 26.3macOS 26.3
Archarm64arm64
Power Modebalancedbalanced
Performance
qwen3.5-0.8b-mlx qwen2.5-math-1.5b-…
Tokens/sec68.291.7
First chunk3268 ms205 ms
TTFT3.3 s205 ms
Load timeN/AN/A
Memory usage0.6 GB0.8 GB
Memory %2%3%
HW Fit Score Breakdown
qwen3.5-0.8b-mlx
Speed
50/50
TTFT
13/20
Memory
24/30
qwen2.5-math-1.5b-instr…
Speed
50/50
TTFT
20/20
Memory
29/30
Quality
qwen3.5-0.8b-mlx
Reasoning
8/20
Coding
1/20
Instruction
12/20
Structured
12/15
Math
3/15
Multilingual
5/10
Reasoning: Weak Coding: Poor Instruction Following: Adequate Structured Output: Strong Math: Poor Multilingual: Adequate
qwen2.5-math-1.5b-instr…
Reasoning
2/20
Coding
0/20
Instruction
3/20
Structured
1/15
Math
14/15
Multilingual
2/10
Reasoning: Poor Coding: Poor Instruction Following: Poor Structured Output: Poor Math: Strong Multilingual: Poor
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