Meta Llama 3.1 405B Instruct vs MiniMax MiniMax M2

Detailed comparison for LLMs

MetaMiniMax

On Guardion's LLM vulnerability Benchmark, Meta Llama 3.1 405B Instruct is the more secure of the two: Llama 3.1 405B Instruct scores 36.0% and MiniMax M2 scores 40.0% 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

Llama 3.1 405B Instruct is the overall winner in this comparison!

Attack Success Rate (lower is safer)

ASR for Meta Llama 3.1 405B Instruct vs MiniMax MiniMax M2. Green marks the safer model on each metric. Only the overall score is available for estimated models.

Overall (ASR)

Llama 3.1 405B Instruct
36.0%
MiniMax M2
40.0%

TAP Attack Method (ASR)

Llama 3.1 405B Instruct
100.0%
MiniMax M2
100.0%

Crescendo Attack Method (ASR)

Llama 3.1 405B Instruct
100.0%
MiniMax M2
100.0%

Zero-Shot (ASR)

Llama 3.1 405B Instruct
100.0%
MiniMax M2
100.0%

Key Highlights

  • Meta Llama 3.1 405B Instruct has a lower Overall (ASR).

Security Profile

Outward is better on every axis.

OverallTAPCrescendoZero-Shot
Llama 3.1 405B Instruct
MiniMax M2
Full security profile
Meta Llama 3.1 405B Instruct
Full security profile
MiniMax MiniMax M2

Frequently asked questions

Is Meta Llama 3.1 405B Instruct or MiniMax MiniMax M2 more secure?

On Guardion's LLM vulnerability Benchmark, Meta Llama 3.1 405B Instruct is the more secure of the two: Llama 3.1 405B Instruct scores 36.0% and MiniMax M2 scores 40.0% 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 Llama 3.1 405B Instruct vs MiniMax M2?

Llama 3.1 405B Instruct has a 36.0% ASR and MiniMax M2 has a 40.0% ASR — the share of adversarial prompts that succeed across zero-shot, TAP, and Crescendo attacks. Lower is safer.

How were Llama 3.1 405B Instruct and MiniMax M2 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