Alibaba Qwen2 7B Instruct vs Alibaba Qwen 2.5 72B

Detailed comparison for LLMs

AlibabaAlibaba

On Guardion's LLM vulnerability Benchmark, Alibaba Qwen 2.5 72B is the more secure of the two: Qwen2 7B Instruct scores 52.0% and Qwen 2.5 72B scores 26.3% 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

Qwen 2.5 72B is the overall winner in this comparison!

Attack Success Rate (lower is safer)

ASR for Alibaba Qwen2 7B Instruct vs Alibaba Qwen 2.5 72B. Green marks the safer model on each metric. Only the overall score is available for estimated models.

Overall (ASR)

Qwen2 7B Instruct
52.0%
Qwen 2.5 72B
26.3%

TAP Attack Method (ASR)

Qwen2 7B Instruct
100.0%
Qwen 2.5 72B
55.2%

Crescendo Attack Method (ASR)

Qwen2 7B Instruct
100.0%
Qwen 2.5 72B
23.7%

Zero-Shot (ASR)

Qwen2 7B Instruct
100.0%
Qwen 2.5 72B
0.3%

Key Highlights

  • Alibaba Qwen 2.5 72B has a lower Overall (ASR).
  • Alibaba Qwen 2.5 72B has a lower TAP Attack Method (ASR).
  • Alibaba Qwen 2.5 72B has a lower Crescendo Attack Method (ASR).
  • Alibaba Qwen 2.5 72B has a lower Zero-Shot (ASR).

Security Profile

Outward is better on every axis.

OverallTAPCrescendoZero-Shot
Qwen2 7B Instruct
Qwen 2.5 72B
Full security profile
Alibaba Qwen2 7B Instruct
Full security profile
Alibaba Qwen 2.5 72B

Frequently asked questions

Is Alibaba Qwen2 7B Instruct or Alibaba Qwen 2.5 72B more secure?

On Guardion's LLM vulnerability Benchmark, Alibaba Qwen 2.5 72B is the more secure of the two: Qwen2 7B Instruct scores 52.0% and Qwen 2.5 72B scores 26.3% 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 Qwen2 7B Instruct vs Qwen 2.5 72B?

Qwen2 7B Instruct has a 52.0% ASR and Qwen 2.5 72B has a 26.3% ASR — the share of adversarial prompts that succeed across zero-shot, TAP, and Crescendo attacks. Lower is safer.

How were Qwen2 7B Instruct and Qwen 2.5 72B 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