In brief: Cisco’s Antares model family enables security teams to analyse known vulnerabilities across 500 entries in approximately 15 minutes on a single GPU for under one US dollar — locally and without data transfer to the cloud.
Cisco has released Antares-350M and Antares-1B, two compact small language models specifically designed to support security teams in analysing known security gaps in source code. The open-weight models run locally and promise lower costs than established solutions.
Cisco provides two small language models — Antares-350M and Antares-1B — specifically trained to identify known security vulnerabilities in source code. The models first analyse the description of a vulnerability, then search for matching code patterns and iteratively adapt their search strategy to new findings. This approach differs from classical static code analysis tools, which typically deliver large quantities of potential matches, and results in faster, more traceable and reproducible assessments.
The economic dimension is substantial for the CISO target audience: a complete analysis of 500 vulnerability entries takes approximately 15 minutes on a single GPU and costs less than one US dollar. This makes AI-supported code analysis practical for smaller organisations, government agencies and academic institutions that have previously had to forgo such procedures due to budget constraints. Particularly relevant is the ability to operate locally: source code remains entirely within the organisation’s own IT environment and does not need to be transferred to cloud services — a decisive advantage for organisations with high data protection and compliance requirements.
Cisco positions Antares for integration into various security processes: identifying vulnerable files in software projects, supporting vulnerability assessment, complementing classical static analysis, and in CI/CD pipelines for early detection of at-risk components. Crucially, Cisco clarifies: Antares neither replaces manual analysis nor a comprehensive application security platform. Complementary procedures such as dependency analysis, secret scanning, dynamic testing, container and infrastructure checks, and expert technical assessment remain indispensable.
Cisco supplements the model family with a new “Vulnerability Localization Benchmark”, open specifications for agent-based security assessments and additional programming guidelines. The goal is to promote standardised AI tools for cybersecurity.
Source: www.it-daily.net · Published 22 July 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.