lfm2.5-thinking:1.2b
OLLAMA GGUFlfm2 · 1.2B · Q4_K_M
May 25, 2026 · Apple M4 Pro
qwen3:4b
OLLAMA GGUFqwen3 · 4.0B · Q4_K_M
Jun 26, 2026 · Cortex-A78AE
Global Score
60 vs 29
Hardware Fit
80 vs 86
Quality Score
52 vs 4
Hardware
lfm2.5-thinking:1.2b qwen3:4b
MachineMacBook ProNVIDIA NVIDIA Jetson Orin Nano Engineering Reference Developer Kit Super
CPUApple M4 ProCortex-A78AE
Cores146
RAM48 GB7 GB
GPUApple M4 ProIntegrated / Unknown
OSmacOS 26.4.1Ubuntu 22.04.5 LTS
Archarm64arm64
Power Modebalancedbalanced
Performance
lfm2.5-thinking:1.2b qwen3:4b
Tokens/sec227.017.9
First chunk60 ms1010 ms
TTFT30.0 s6.4 s
Load time0.7 s9.5 s
Memory usage2.6 GB3.0 GB
Memory %6%40%
HW Fit Score Breakdown
lfm2.5-thinking:1.2b
Speed
50/50
TTFT
0/20
Memory
30/30
qwen3:4b
Speed
50/50
TTFT
8/20
Memory
28/30
Quality
lfm2.5-thinking:1.2b
Reasoning
8/20
Coding
6/20
Instruction
10/20
Structured
7/15
Math
12/15
Multilingual
9/10
Reasoning: Weak Coding: Weak Instruction Following: Adequate Structured Output: Weak Math: Strong Multilingual: Strong
qwen3:4b
Reasoning
4/20
Coding
0/20
Instruction
0/20
Structured
0/15
Math
0/15
Multilingual
0/10
Reasoning: Poor Coding: Poor Instruction Following: Poor Structured Output: Poor Math: Poor 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