Cisco’s open-weight Antares models make vulnerability localization cheaper

Cisco announced Antares, a family of compact language models (Antares-350M, Antares-1B, with Antares-3B forthcoming) that guide analysts to source files likely to contain known vulnerabilities; the models operate by searching, reading, backtracking, and ranking candidate files, and are evaluated on a new Vulnerability Localization Benchmark where they show strong accuracy and much lower cost and runtime compared with larger models.