The gap between open and closed AI models in cyber capabilities keeps shrinking
The Decoder
Open-weight AI models are closing the capability gap with closed models in simulated cyberattacks while remaining significantly cheaper.
According to the Vals AI benchmark, OpenAI's GPT-6 Sol Daybreak successfully achieved 80.8 percent of attack objectives in simulated corporate networks. Meanwhile, the open-weight model DeepSeek V4.1 Flash reached 72.8 percent on the same benchmark at a fraction of the cost. Analysts point out that because open models cannot be easily restricted, the window of time for defenders is rapidly shrinking.
Why it matters: The rapid advancement and low cost of capable open-weight cyber AI models mean that defenders have a quickly closing window to protect corporate networks.






