Cisco's Antares AI Models Scan Code for Security Bugs at Under $1 Per Run

Cisco Foundation AI has released two open-weight language models, Antares-350M and Antares-1B, specifically designed to detect security vulnerabilities in codebases by linking CVE entries to affected files. In benchmarks using a 500-entry vulnerability dataset, Antares completed scans in roughly one hour at a cost under $1, compared to 4.5 hours and $141 for OpenAI's GPT-5.5. Both models run locally, ensuring sensitive source code never leaves the user's own hardware. Cisco has also announced a larger Antares-3B model is in development, and the existing models are publicly available on Hugging Face. The release highlights a growing trend of purpose-built, small-scale AI models that prioritise cost efficiency and privacy over broad general capability.