Back to leaderboard

qwen3-30b-a3b-thinking-2507-claude-4.5-sonnet-high-reasoning-distill-mlx

LM-STUDIO MLX

qwen3_moe · 30B · MXFP4

Excellent

Apr 22, 2026 · Apple M2 Max

openai/gpt-oss-20b

LM-STUDIO MLX

gpt_oss · 20B · MXFP4

Not Rec.

Mar 19, 2026 · Apple M4

Global Score
95 vs 77
Hardware Fit
94 vs 67
Quality Score
96 vs 81

Hardware

qwen3-30b-a3b-thin… openai/gpt-oss-20b
MachineMac StudioMac mini
CPUApple M2 MaxApple M4
Cores1210
RAM96 GB24 GB
GPUApple M2 MaxApple M4
OSmacOS 26.1macOS 26.3
Archarm64arm64
Power Modebalancedbalanced

Performance

qwen3-30b-a3b-thin… openai/gpt-oss-20b
Tokens/sec80.430.5
First chunk46 ms18 ms
TTFT1.8 s30.0 s
Load timeN/AN/A
Memory usage21.2 GB15.8 GB
Memory %22%66%

HW Fit Score Breakdown

qwen3-30b-a3b-thinking-…

Speed
50/50
TTFT
14/20
Memory
30/30

openai/gpt-oss-20b

Speed
50/50
TTFT
0/20
Memory
17/30

Quality

qwen3-30b-a3b-thinking-…

Reasoning
19/20
Coding
18/20
Instruction
19/20
Structured
15/15
Math
15/15
Multilingual
10/10
Reasoning: Strong Coding: Strong Instruction Following: Strong Structured Output: Strong Math: Strong Multilingual: Strong

openai/gpt-oss-20b

Reasoning
19/20
Coding
11/20
Instruction
12/20
Structured
15/15
Math
14/15
Multilingual
10/10
Reasoning: Strong Coding: Adequate Instruction Following: Adequate Structured Output: Strong Math: Strong Multilingual: Strong

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