July 2026 at Foundation AI: A Roundup

Cisco's Foundation AI team released Antares, a series of small, open-weight models designed for identifying software vulnerabilities. These models, available in 350 million, 1 billion, and 3 billion parameter sizes, focus specifically on agentic vulnerability localization.
Key takeaways
- New open-weight models named Antares released
- Models specialize in finding software vulnerabilities
- Compact sizes available: 350M, 1B, and 3B parameters
- Focus on agentic vulnerability localization
Why it matters
Developers and security professionals can leverage these specialized models to automate parts of the vulnerability discovery process. This could lead to faster identification and patching of security flaws in software, improving overall system security.


