Cisco Foundation AI has unveiled Antares, a new family of open-weight language models designed to identify known vulnerabilities within real-world codebases. The release includes two models: Antares-350M and Antares-1B, both tailored for the specific task of vulnerability localization. These models mark a significant advancement in AI-assisted cybersecurity, offering both efficiency and accuracy in pinpointing potential security flaws.
Performance and Efficiency
The Antares-1B model, with 1 billion parameters, achieves a File F1 score of 0.209 on a newly introduced Vulnerability Localization Benchmark. This performance surpasses that of larger models like GLM-5.2, which has 753 billion parameters, and even Google’s Gemini 3 Pro. Notably, the models’ capabilities are largely derived from post-training, as untrained models such as Granite 4.0 checkpoints perform nearly at zero under the same evaluation protocol.
One of Antares’ standout features is its efficiency. The full 500-task evaluation runs in approximately 13 minutes on a single NVIDIA H100 GPU and costs less than a dollar—compared to the $141 required for a similar task using GPT-5.5. This cost-effective and fast performance positions Antares as a practical tool for developers and security teams looking to integrate vulnerability detection into their workflows.
Implications for Cybersecurity
Antares addresses a critical need in software security: the ability to quickly and accurately locate known vulnerabilities in code. As software systems grow increasingly complex, manual code reviews become less feasible, and AI-driven tools like Antares can help automate and enhance the process. The open-weight nature of the models also allows for broader adoption and customization by the developer community.
This release underscores the growing trend of specialized AI models for cybersecurity tasks. By focusing on a narrow but vital domain—vulnerability localization—Cisco Foundation AI has created a tool that not only performs well but also offers practical advantages in terms of cost and speed. Antares could play a pivotal role in strengthening software security practices across industries.
