Google Gemma 3n E4B vs Google Gemma 3n E4B Instructed

Detailed comparison for LLMs

GoogleGoogle

On Guardion's LLM vulnerability Benchmark, Google Gemma 3n E4B Instructed is the more secure of the two: Gemma 3n E4B scores 52.0% and Gemma 3n E4B Instructed scores 49.0% 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

Gemma 3n E4B Instructed is the overall winner in this comparison!

Attack Success Rate (lower is safer)

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

Overall (ASR)

Gemma 3n E4B
52.0%
Gemma 3n E4B Instructed
49.0%

TAP Attack Method (ASR)

Gemma 3n E4B
100.0%
Gemma 3n E4B Instructed
100.0%

Crescendo Attack Method (ASR)

Gemma 3n E4B
100.0%
Gemma 3n E4B Instructed
100.0%

Zero-Shot (ASR)

Gemma 3n E4B
100.0%
Gemma 3n E4B Instructed
100.0%

Key Highlights

  • Google Gemma 3n E4B Instructed has a lower Overall (ASR).

Security Profile

Outward is better on every axis.

OverallTAPCrescendoZero-Shot
Gemma 3n E4B
Gemma 3n E4B Instructed
Full security profile
Google Gemma 3n E4B
Full security profile
Google Gemma 3n E4B Instructed

Frequently asked questions

Is Google Gemma 3n E4B or Google Gemma 3n E4B Instructed more secure?

On Guardion's LLM vulnerability Benchmark, Google Gemma 3n E4B Instructed is the more secure of the two: Gemma 3n E4B scores 52.0% and Gemma 3n E4B Instructed scores 49.0% 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 Gemma 3n E4B vs Gemma 3n E4B Instructed?

Gemma 3n E4B has a 52.0% ASR and Gemma 3n E4B Instructed has a 49.0% ASR — the share of adversarial prompts that succeed across zero-shot, TAP, and Crescendo attacks. Lower is safer.

How were Gemma 3n E4B and Gemma 3n E4B Instructed 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