OpenAI GPT OSS 120B vs Anthropic Claude 4.0 Sonnet

Detailed comparison for LLMs

OpenAIAnthropic

On Guardion's LLM vulnerability Benchmark, Anthropic Claude 4.0 Sonnet is the more secure of the two: GPT OSS 120B scores 15.0% and Claude 4.0 Sonnet scores 13.6% 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 4.0 Sonnet is the overall winner in this comparison!

Attack Success Rate (lower is safer)

ASR for OpenAI GPT OSS 120B vs Anthropic Claude 4.0 Sonnet. Green marks the safer model on each metric. Only the overall score is available for estimated models.

Overall (ASR)

GPT OSS 120B
15.0%
Claude 4.0 Sonnet
13.6%

TAP Attack Method (ASR)

GPT OSS 120B
100.0%
Claude 4.0 Sonnet
22.0%

Crescendo Attack Method (ASR)

GPT OSS 120B
100.0%
Claude 4.0 Sonnet
14.8%

Zero-Shot (ASR)

GPT OSS 120B
100.0%
Claude 4.0 Sonnet
4.1%

Key Highlights

  • Anthropic Claude 4.0 Sonnet has a lower Overall (ASR).
  • Anthropic Claude 4.0 Sonnet has a lower TAP Attack Method (ASR).
  • Anthropic Claude 4.0 Sonnet has a lower Crescendo Attack Method (ASR).
  • Anthropic Claude 4.0 Sonnet has a lower Zero-Shot (ASR).

Security Profile

Outward is better on every axis.

OverallTAPCrescendoZero-Shot
GPT OSS 120B
Claude 4.0 Sonnet
Full security profile
OpenAI GPT OSS 120B
Full security profile
Anthropic Claude 4.0 Sonnet

Frequently asked questions

Is OpenAI GPT OSS 120B or Anthropic Claude 4.0 Sonnet more secure?

On Guardion's LLM vulnerability Benchmark, Anthropic Claude 4.0 Sonnet is the more secure of the two: GPT OSS 120B scores 15.0% and Claude 4.0 Sonnet scores 13.6% 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 GPT OSS 120B vs Claude 4.0 Sonnet?

GPT OSS 120B has a 15.0% ASR and Claude 4.0 Sonnet has a 13.6% ASR — the share of adversarial prompts that succeed across zero-shot, TAP, and Crescendo attacks. Lower is safer.

How were GPT OSS 120B and Claude 4.0 Sonnet 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