Mistral Mixtral 8x7B vs Mistral Mistral Large 3 675B Instruct 2512 NVFP4
Detailed comparison for LLMs
On Guardion's LLM vulnerability Benchmark, Mistral Mistral Large 3 675B Instruct 2512 NVFP4 is the more secure of the two: Mixtral 8x7B scores 38.0% and Mistral Large 3 675B Instruct 2512 NVFP4 scores 34.0% 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
Mistral Large 3 675B Instruct 2512 NVFP4 is the overall winner in this comparison!
Attack Success Rate (lower is safer)
ASR for Mistral Mixtral 8x7B vs Mistral Mistral Large 3 675B Instruct 2512 NVFP4. 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
- Mistral Mistral Large 3 675B Instruct 2512 NVFP4 has a lower Overall (ASR).
- Mistral Mixtral 8x7B has a lower TAP Attack Method (ASR).
- Mistral Mixtral 8x7B has a lower Crescendo Attack Method (ASR).
- Mistral Mixtral 8x7B has a lower Zero-Shot (ASR).
Security Profile
Outward is better on every axis.
Frequently asked questions
Is Mistral Mixtral 8x7B or Mistral Mistral Large 3 675B Instruct 2512 NVFP4 more secure?
On Guardion's LLM vulnerability Benchmark, Mistral Mistral Large 3 675B Instruct 2512 NVFP4 is the more secure of the two: Mixtral 8x7B scores 38.0% and Mistral Large 3 675B Instruct 2512 NVFP4 scores 34.0% 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 Mixtral 8x7B vs Mistral Large 3 675B Instruct 2512 NVFP4?
Mixtral 8x7B has a 38.0% ASR and Mistral Large 3 675B Instruct 2512 NVFP4 has a 34.0% ASR — the share of adversarial prompts that succeed across zero-shot, TAP, and Crescendo attacks. Lower is safer.
How were Mixtral 8x7B and Mistral Large 3 675B Instruct 2512 NVFP4 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.