deepseek-v2:16b

deepseek2 · 15.7B · Q4_0

Good

Mar 2, 2026 · Apple M4

llama3.2:3b

llama · 3.2B · Q4_K_M

Good

Mar 1, 2026 · Apple M4

Global Score

73 vs 69

Hardware Fit

100 vs 98

Quality Score

55 vs 49

Hardware

deepseek-v2:16b llama3.2:3b

MachineMacBook AirMacBook Air

CPUApple M4Apple M4

Cores1010

RAM32 GB32 GB

GPUApple M4Apple M4

OSmacOS 26.3macOS 26.3

Archarm64arm64

Power Modebalancedunknown

Performance

deepseek-v2:16b llama3.2:3b

Tokens/sec56.627.8

TTFT239 ms285 ms

Load time1.2 s1.9 s

Memory usage17.7 GB7.0 GB

Memory %55%22%

HW Fit Score Breakdown

deepseek-v2:16b

Speed

40/40

TTFT

30/30

Memory

30/30

llama3.2:3b

Speed

38/40

TTFT

30/30

Memory

30/30

Quality

deepseek-v2:16b

Reasoning

7/20

Coding

9/20

Instruction

13/20

Structured

10/15

Math

8/15

Multilingual

8/10

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

llama3.2:3b

Reasoning

11/20

Coding

0/20

Instruction

14/20

Structured

12/15

Math

3/15

Multilingual

9/10

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

View deepseek-v2:16b details | View llama3.2:3b details

Run yours and compare

npx metrillm@latest bench

Requires Node 20+ and Ollama or LM Studio running