Trusted-Access Models Added to the Artificial Analysis Cyber Index

Trusted-Access Models Added to the Artificial Analysis Cyber Index

GPT-6 Sol (Daybreak Blue) now leads the Cyber Index.

The Artificial Analysis Cyber Index Alliance evaluates how AI models perform on enterprise cyber defense tasks. When the Index launched, it included only publicly accessible models, providing a view of publicly available capability in agentic cyber defense.

The leaderboard is now expanding to include trusted-access models with fewer cyber guardrails. This change is intended to provide a broader view of frontier-model capability in cyber defense. The first addition is GPT-6 Sol (Daybreak Blue, max), which is accessible only through OpenAI’s Daybreak program.

GPT-6 Sol (Daybreak Blue) leads the Index

GPT-6 Sol (Daybreak Blue, max) now leads the Cyber Index and sits on the Cost vs. Capability Pareto frontier. It encountered no safety blocks across the Index, and its overall score is 32 points higher than that of the publicly available GPT-6 Sol (max).

With a Cost per Task of $1.77, it is less expensive per task than other leading models, including Grok 4.7 (xhigh), which costs $11.67 per task.

Compared with the publicly available GPT-6 Sol, Daybreak Blue shows its largest gains on CyberGym-E2E. This is also the benchmark where the Index records the most safety refusals.

Artificial Analysis plans to continue expanding the Cyber Index to cover both publicly available and trusted-access models. Full evaluation results are available on the Artificial Analysis Cyber Index evaluation page.