qwen3:4b
OLLAMA GGUFqwen3 · 4.0B · Q4_K_M
Jun 26, 2026 · Cortex-A78AE
qwen3:14b
OLLAMA GGUFqwen3 · 14.8B · Q4_K_M
Jun 13, 2026 · AMD Ryzen 9 3900X 12-Core Processor
Global Score
29 vs 82
Hardware Fit
86 vs 90
Quality Score
4 vs 79
Hardware
qwen3:4b qwen3:14b
MachineNVIDIA NVIDIA Jetson Orin Nano Engineering Reference Developer Kit SuperMicro-Star International Co., Ltd. MS-7C37
CPUCortex-A78AEAMD Ryzen 9 3900X 12-Core Processor
Cores624
RAM7 GB64 GB
GPUIntegrated / UnknownGameViewer Virtual Display Adapter, NVIDIA GeForce RTX 3080
OSUbuntu 22.04.5 LTSMicrosoft Windows 11 专业版 10.0.26200
Archarm64x64
Power Modebalancedbalanced
Performance
qwen3:4b qwen3:14b
Tokens/sec17.974.4
First chunk1010 ms324 ms
TTFT6.4 s2.6 s
Load time9.5 s5.1 s
Memory usage3.0 GB9.0 GB
Memory %40%14%
HW Fit Score Breakdown
qwen3:4b
Speed
50/50
TTFT
8/20
Memory
28/30
qwen3:14b
Speed
50/50
TTFT
10/20
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
qwen3:14b
Reasoning
18/20
Coding
11/20
Instruction
17/20
Structured
14/15
Math
10/15
Multilingual
9/10
Reasoning: Strong Coding: Adequate Instruction Following: Strong Structured Output: Strong 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