qwen3:4b
OLLAMA GGUFqwen3 · 4.0B · Q4_K_M
Jun 26, 2026 · Cortex-A78AE
lfm2.5-thinking:1.2b
OLLAMA GGUFlfm2 · 1.2B · Q4_K_M
May 25, 2026 · Apple M4 Pro
Global Score
32 vs 60
Hardware Fit
98 vs 80
Quality Score
4 vs 52
Hardware
qwen3:4b lfm2.5-thinking:1.2b
MachineNVIDIA NVIDIA Jetson Orin Nano Engineering Reference Developer Kit SuperMacBook Pro
CPUCortex-A78AEApple M4 Pro
Cores614
RAM7 GB48 GB
GPUIntegrated / UnknownApple M4 Pro
OSUbuntu 22.04.5 LTSmacOS 26.4.1
Archarm64arm64
Power Modebalancedbalanced
Performance
qwen3:4b lfm2.5-thinking:1.2b
Tokens/sec17.9227.0
First chunk950 ms60 ms
TTFT950 ms30.0 s
Load time9.7 s0.7 s
Memory usage3.0 GB2.6 GB
Memory %40%6%
HW Fit Score Breakdown
qwen3:4b
Speed
50/50
TTFT
20/20
Memory
28/30
lfm2.5-thinking:1.2b
Speed
50/50
TTFT
0/20
Memory
30/30
Quality
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
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
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