Deepseek Deepseek V3 vs Google Gemma 3n E4B

Detailed comparison for LLMs

DeepseekGoogle

On Guardion's LLM vulnerability Benchmark, Deepseek Deepseek V3 is the more secure of the two: Deepseek V3 scores 47.2% and Gemma 3n E4B scores 52.0% on attack success rate (ASR) (lower is better). One or both scores are estimated from public safety evaluations pending a Guardion benchmark run.

Head-to-Head Overview

Deepseek V3 is the overall winner in this comparison!

Attack Success Rate (lower is safer)

ASR for Deepseek Deepseek V3 vs Google Gemma 3n E4B. Green marks the safer model on each metric. Only the overall score is available for estimated models.

Overall (ASR)

Deepseek V3
47.2%
Gemma 3n E4B
52.0%

TAP Attack Method (ASR)

Deepseek V3
79.5%
Gemma 3n E4B
100.0%

Crescendo Attack Method (ASR)

Deepseek V3
49.3%
Gemma 3n E4B
100.0%

Zero-Shot (ASR)

Deepseek V3
12.8%
Gemma 3n E4B
100.0%

Key Highlights

  • Deepseek Deepseek V3 has a lower Overall (ASR).
  • Deepseek Deepseek V3 has a lower TAP Attack Method (ASR).
  • Deepseek Deepseek V3 has a lower Crescendo Attack Method (ASR).
  • Deepseek Deepseek V3 has a lower Zero-Shot (ASR).

Security Profile

Outward is better on every axis.

OverallTAPCrescendoZero-Shot
Deepseek V3
Gemma 3n E4B
Full security profile
Deepseek Deepseek V3
Full security profile
Google Gemma 3n E4B

Frequently asked questions

Is Deepseek Deepseek V3 or Google Gemma 3n E4B more secure?

On Guardion's LLM vulnerability Benchmark, Deepseek Deepseek V3 is the more secure of the two: Deepseek V3 scores 47.2% and Gemma 3n E4B scores 52.0% on attack success rate (ASR) (lower is better). One or both scores are estimated from public safety evaluations pending a Guardion benchmark run.

What is the attack success rate (ASR) of Deepseek V3 vs Gemma 3n E4B?

Deepseek V3 has a 47.2% ASR and Gemma 3n E4B has a 52.0% ASR — the share of adversarial prompts that succeed across zero-shot, TAP, and Crescendo attacks. Lower is safer.

How were Deepseek V3 and Gemma 3n E4B tested?

Both were red-teamed with the HarmBench framework across zero-shot, TAP (Tree of Attacks with Pruning), and Crescendo multi-turn attacks, scored by Attack Success Rate.

Related Comparisons