deepseek-v2:16b

deepseek2 · 15.7B · Q4_0

Good

Mar 2, 2026 · Apple M4

wizardlm2:7b

llama · 7B · Q4_0

Marginal

Mar 2, 2026 · Apple M4

Global Score

69 vs 59

Hardware Fit

100 vs 97

Quality Score

55 vs 43

Hardware

deepseek-v2:16b wizardlm2:7b

MachineMacBook AirMacBook Air

CPUApple M4Apple M4

Cores1010

RAM32 GB32 GB

GPUApple M4Apple M4

OSmacOS 26.3macOS 26.3

Archarm64arm64

Power Modebalancedbalanced

Performance

deepseek-v2:16b wizardlm2:7b

Tokens/sec56.623.2

TTFT239 ms264 ms

Load time1.2 s0.6 s

Memory usage17.7 GB9.9 GB

Memory %55%31%

HW Fit Score Breakdown

deepseek-v2:16b

Speed

40/40

TTFT

30/30

Memory

30/30

wizardlm2:7b

Speed

37/40

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

wizardlm2:7b

Reasoning

6/20

Coding

11/20

Instruction

10/20

Structured

4/15

Math

4/15

Multilingual

8/10

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

View deepseek-v2:16b details | View wizardlm2:7b details

Run yours and compare

npx metrillm@latest bench

Requires Node 20+ and Ollama or LM Studio running