baidu/ernie-4.5-21b-a3b

LM-STUDIO GGUF

ernie4_5-moe · 21B · Q4_K_M

Excellent

Nov 1, 2026 · AMD RYZEN AI MAX+ 395

mistralai/devstral-small-2-2512

LM-STUDIO MLX

mistral3 · 24B · 4bit

Good

Oct 20, 2026 · Apple M1 Pro

Global Score

86 vs 73

Hardware Fit

93 vs 64

Quality Score

83 vs 77

Hardware

baidu/ernie-4.5-21… mistralai/devstral…

MachineGMKtec EVO-X3MacBook Pro

CPUAMD RYZEN AI MAX+ 395Apple M1 Pro

Cores3210

RAM120 GB32 GB

GPUAMD Radeon(TM) 8060S GraphicsApple M1 Pro

OSMicrosoft Windows 11 Pro 10.0.26300macOS 15.6.1

Archx64arm64

Power Modebalancedbalanced

Performance

baidu/ernie-4.5-21… mistralai/devstral…

Tokens/sec46.311.9

First chunk33 ms7 ms

TTFT383 ms1.1 s

Load time49.7 sN/A

Memory usage32.5 GB19.7 GB

Memory %27%62%

HW Fit Score Breakdown

baidu/ernie-4.5-21b-a3b

Speed

43/50

TTFT

20/20

Memory

30/30

mistralai/devstral-smal…

Speed

25/50

TTFT

20/20

Memory

19/30

Quality

baidu/ernie-4.5-21b-a3b

Reasoning

16/20

Coding

17/20

Instruction

15/20

Structured

14/15

Math

12/15

Multilingual

9/10

Reasoning: Strong Coding: Strong Instruction Following: Strong Structured Output: Strong Math: Strong Multilingual: Strong

mistralai/devstral-smal…

Reasoning

17/20

Coding

16/20

Instruction

13/20

Structured

14/15

Math

7/15

Multilingual

10/10

Reasoning: Strong Coding: Strong Instruction Following: Adequate Structured Output: Strong Math: Weak Multilingual: Strong

View baidu/ernie-4.5-21… details | View mistralai/devstral… 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