0
Applied AI·July 22, 2026·1 min read

Cisco’s tiny open-weight AI hunts bugs, and it says it beats Gemini and GPT

Share

If Cisco’s Antares models really outperform frontier LLMs on vulnerability discovery while staying small and open-weight, security teams now have a credible path to run high-value AI locally instead of piping code through external APIs. The gating of access is the tell—treat these models like dual-use tools and update your internal policies on who can run automated bug-hunting and where the outputs are stored.