Foundation AI in July: New Models, Benchmarks, and Security Research
Cisco, Wednesday, August 19th, 2026
Cisco's Foundation AI team recaps July, headlined by the Antares family of compact open-weight vulnerability-localization models.
Cisco's Foundation AI team published its July recap, which it describes as an Antares month with a parallel thread on smarter evaluation.
Antares is a family of compact, open-weight models at 350M, 1B and 3B parameters built for a single job: agentic vulnerability localization.
The technical report accompanying the release documents how the models were built and how they perform. The post also covers new benchmarks the team shipped and its security research output for the month. The through-line is that small task-specific models can outperform general ones on narrow security tasks.