Foundation AI in July: New Models, Benchmarks, and Security Research

Cisco releases Antares, a family of compact open-weight models designed for agentic vulnerability localization in codebases. The 350M and 1B variants are publicly available on Hugging Face, alongside a CLI tool that generates SARIF reports for GitHub Code Scanning. The system claims to rival frontier models in pinpointing vulnerabilities while running locally, offering a privacy-preserving alternative for security workflows.