OpenAI's lower-guardrail GPT-6 Sol tops Cyber Index

Artificial Analysis has expanded its Cyber Index to include models that run with fewer cyber guardrails. The first addition, GPT-6 Sol (Daybreak Blue, max), now leads the leaderboard. The model is available only through OpenAI's Daybreak program.

The Cyber Index is run through the Artificial Analysis Cyber Index Alliance, a group of industry partners that evaluates how AI models perform on enterprise cyber defense tasks. Until now it included only publicly accessible models, so it reflected the agentic cyber defense capability anyone could get. Adding trusted-access models is meant to give "a broader view of frontier model cyber capability," according to Artificial Analysis.

According to the published results:

• Score: Daybreak Blue's overall score is 32 points higher than the publicly available GPT-6 Sol (max).
• Refusals: It hit no safety blocks anywhere in the Index.
• Cost: At $1.77 per task, it sits on the Index's cost-vs-capability Pareto frontier. It costs less than other leading models, including Grok 4.7 (xhigh) at $11.67 per task.
• Biggest gain: It improved most on CyberGym-E2E, the benchmark where Artificial Analysis sees the most safety refusals.

That pattern suggests much of the gap between the two GPT-6 Sol variants comes from fewer refusals rather than different underlying capability. The summary does not separate the two effects, and it does not give absolute scores or say what the Daybreak program requires for access.

Artificial Analysis said it will keep covering both publicly available and trusted-access models in the Index. Full results are on its evaluation page.