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

Cisco has developed small, open-weight AI models named Antares-350M and Antares-1B that detect software vulnerabilities efficiently and cost-effectively, outperforming larger models like Google's Gemini and OpenAI's GPT on certain benchmarks. These models run locally, preserving code privacy, and Cisco controls access to them, with a larger 3-billion-parameter model withheld from full public release.