Antares: Cisco's Open-Weight AI for Vulnerability Localization

Cisco is introducing Antares, a new family of security small language models. These models are designed to find known vulnerabilities within a codebase, a task that is often difficult and expensive. Two of these models, Antares-350M and Antares-1B, are now available as open-weight models on Hugging Face. Benchmark tests show they outperform many other models for this security task, at a lower cost. What's interesting is they are compact enough to run locally, so sensitive code doesn't need to go to the cloud. Cisco is combining these open-weight models with other tools and standards to help cybersecurity professionals. The bottom line is Antares aims to provide practical, trustworthy AI tools for better security.