Claude Mythos 5.276.3%
Anthropic 10-challenge subset / prerelease10 private challenges; cyber mitigations off; prerelease snapshot assessed similar to release. Each lab uses different tasks. · Tools enabled
Anthropic ↗Best LLMs for multi-stage security operations
CyScenarioBenchTests multi-stage cybersecurity scenarios, helping authorized security teams assess an agent's ability to plan and execute complex assessments. Labs use different private task sets, and safeguards settings are recorded with each report.
Updated: Data as of:
| Model / creator | Version | Reported score |
|---|---|---|
 Claude Mythos 5.2Anthropic · Proprietary | Anthropic 10-challenge subset / prerelease | 76.3% |
| Claude Opus 5.5Anthropic · Proprietary | Anthropic 10-challenge subset / prerelease | 67.6% |
| Claude Mythos 5.1Anthropic · Proprietary | Anthropic 10-challenge subset / prerelease | 61.7% |
 GPT-6 CyberOpenAI · Proprietary | OpenAI GPT-6 Cyber / unique task set | 61% |
View all 4 published reports
Claude Opus 5.567.6%
Anthropic 10-challenge subset / prereleaseeffort: highest available · safeguards: false · November 9 launch-table comparison; highest available effort unless explicitly noted. Preserve the benchmark track and model configuration; competitor figures are as reported by Anthropic. · Tools enabled
Anthropic ↗Claude Mythos 5.161.7%
Anthropic 10-challenge subset / prereleaseeffort: highest available · safeguards: false · November 9 launch-table comparison; highest available effort unless explicitly noted. Preserve the benchmark track and model configuration; competitor figures are as reported by Anthropic. · Tools enabled
Anthropic ↗GPT-6 Cyber61%
OpenAI GPT-6 Cyber / unique task seteffort: highest available · GPT-6 Cyber: 61%. Labs use unique tasks; not a shared task set with the Anthropic 10-challenge subset. · Tools enabled
Anthropic ↗