Anthropic Claude 3.5 Sonnet v2 vs Mistral Mistral Large 3 675B Instruct 2512 Eagle
Detailed comparison for LLMs
On Guardion's LLM vulnerability Benchmark, Anthropic Claude 3.5 Sonnet v2 is the more secure of the two: Claude 3.5 Sonnet v2 scores 4.4% and Mistral Large 3 675B Instruct 2512 Eagle scores 33.0% 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
Claude 3.5 Sonnet v2 is the overall winner in this comparison!
Attack Success Rate (lower is safer)
ASR for Anthropic Claude 3.5 Sonnet v2 vs Mistral Mistral Large 3 675B Instruct 2512 Eagle. Green marks the safer model on each metric. Only the overall score is available for estimated models.
Overall (ASR)
TAP Attack Method (ASR)
Crescendo Attack Method (ASR)
Zero-Shot (ASR)
Key Highlights
- Anthropic Claude 3.5 Sonnet v2 has a lower Overall (ASR).
- Anthropic Claude 3.5 Sonnet v2 has a lower TAP Attack Method (ASR).
- Anthropic Claude 3.5 Sonnet v2 has a lower Crescendo Attack Method (ASR).
- Anthropic Claude 3.5 Sonnet v2 has a lower Zero-Shot (ASR).
Security Profile
Outward is better on every axis.
Frequently asked questions
Is Anthropic Claude 3.5 Sonnet v2 or Mistral Mistral Large 3 675B Instruct 2512 Eagle more secure?
On Guardion's LLM vulnerability Benchmark, Anthropic Claude 3.5 Sonnet v2 is the more secure of the two: Claude 3.5 Sonnet v2 scores 4.4% and Mistral Large 3 675B Instruct 2512 Eagle scores 33.0% 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 Claude 3.5 Sonnet v2 vs Mistral Large 3 675B Instruct 2512 Eagle?
Claude 3.5 Sonnet v2 has a 4.4% ASR and Mistral Large 3 675B Instruct 2512 Eagle has a 33.0% ASR — the share of adversarial prompts that succeed across zero-shot, TAP, and Crescendo attacks. Lower is safer.
How were Claude 3.5 Sonnet v2 and Mistral Large 3 675B Instruct 2512 Eagle 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.