OpenAI released two models: Astra for complex projects and Sol for everyday work. Compared with their predecessors, they’re better at multistep tasks, coding, and producing useful documents with fewer tokens. OpenAI The cyber results are the interesting part for me. In one of OpenAI’s internal tests involving vulnerable software, Astra solved 36.4% of targets versus 18.2% for GPT‑6 Astra. Sol reached 25% versus 11.4% for GPT‑6.1 Sol. These were controlled evaluations, with production safeguards disabled, so they aren’t a promise of the same results in a real environment. To me, AI companies have a clear objective. To integrated AI agents into (not) just testing, but also patching, which to me will be the hardest challenge to deal with. Have any of new tried to fix some vulnerabilities with AI?