Back to leaderboard

qwen3:4b

OLLAMA GGUF

qwen3 · 4.0B · Q4_K_M

Not Rec.

Jun 26, 2026 · Cortex-A78AE

qwen2.5-7b-instruct

LM-STUDIO MLX

qwen2 · 7B · 4bit

Excellent

Mar 3, 2026 · Apple M4

Global Score
32 vs 84
Hardware Fit
98 vs 97
Quality Score
4 vs 78

Hardware

qwen3:4b qwen2.5-7b-instruct
MachineNVIDIA NVIDIA Jetson Orin Nano Engineering Reference Developer Kit SuperMacBook Air
CPUCortex-A78AEApple M4
Cores610
RAM7 GB32 GB
GPUIntegrated / UnknownApple M4
OSUbuntu 22.04.5 LTSmacOS 26.3
Archarm64arm64
Power Modebalancedbalanced

Performance

qwen3:4b qwen2.5-7b-instruct
Tokens/sec17.924.4
First chunk950 ms464 ms
TTFT950 ms464 ms
Load time9.7 sN/A
Memory usage3.0 GB4.0 GB
Memory %40%13%

HW Fit Score Breakdown

qwen3:4b

Speed
50/50
TTFT
20/20
Memory
28/30

qwen2.5-7b-instruct

Speed
47/50
TTFT
20/20
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

qwen2.5-7b-instruct

Reasoning
14/20
Coding
17/20
Instruction
15/20
Structured
15/15
Math
8/15
Multilingual
9/10
Reasoning: Adequate Coding: Strong Instruction Following: Strong Structured Output: Strong Math: Adequate Multilingual: Strong
View qwen3:4b details | View qwen2.5-7b-instruct details

Run yours and compare

$ npm install -g metrillm@latest
$ metrillm

Requires Node 20+ and Ollama or LM Studio running

Or run without installing: npx metrillm@latest