OpenAI's 5.6 Sol model achieved a superior score in a hacking evaluation compared to Anthropic's Mythos, which is not yet fully available. The AI Security Institute's analysis indicates the gap between open and closed frontier cyber capabilities has narrowed to 4-7 months, down from 6-10 months in 2025, with advanced capabilities reaching less safeguarded open models faster.
Open-source models are rapidly acquiring frontier hacking capabilities, increasing the speed at which advanced cyber tools become widely accessible.