Cisco Systems, Inc. Releases Antares AI Models for Vulnerability Localization

Cisco Systems, Inc. released Antares, a family of open-weight AI models that assist security teams in isolating potentially vulnerable sections of software repositories. Rather than detecting specific CVEs or generating patches, the models use Common Weakness Enumeration descriptions to provide a ranked list of files most likely to contain a specific class of vulnerability.

The release includes three models with 350 million, 1 billion, and 3 billion parameters to accommodate hardware ranging from resource-constrained systems to workstations. Cisco Systems, Inc. reports that the 3B model performs comparably to GPT-5.5 on internal benchmarks and outperforms larger models from Google, OpenAI, and Meta. To maintain the security of proprietary code, the models support local inference and are available on Hugging Face.

Cisco Systems, Inc. positions Antares as a search assistant intended to reduce investigator fatigue and workload rather than a replacement for human judgment or existing security toolchains. The models output a ranked list of source files and the terminal exploration trace used to reach that result.