Cisco has launched two new small language models focused on cybersecurity, named Antares-350M and Antares-1B, designed to enhance codebase analysis while reducing operational costs. These models will be available as open-weight options through Hugging Face, aiming to provide a more affordable solution for identifying code vulnerabilities compared to existing frontier models.
The Antares models are positioned as specialized tools that will operate alongside both frontier and generalized models, promising to run at a “fraction of the compute expense” associated with traditional models. Amin Karbasi, VP and chief scientist for Cisco Foundation AI, stated, “Benchmark testing shows that these models outperform many powerful closed- and open-weight models in this critical security task at a fraction of the cost.”
Karabasi emphasized the practicality of the Antares models, noting their capability to run locally, which helps eliminate the need to transfer sensitive codebases to the cloud for analysis. This development comes amid growing concerns about rising AI costs across various sectors, leading some companies, such as Accenture, to advise employees to limit AI usage due to escalating expenses.
The Antares models are designed to assist human investigators by streamlining the process of navigating through code repositories. They can help identify files related to vulnerability warnings, initiate investigations based on advisories, and support workflows that involve vulnerable files. Cisco’s decision to release these models as open-weight allows for greater accessibility and adaptability within the security community.
Karbasi noted, “We are releasing Antares as open-weight because the security community needs more accessible building blocks for practical, repository-level defense.” He further explained that Cisco’s approach is rooted in the belief that AI in security must evolve beyond standalone demonstrations and into systems that practitioners can evaluate, govern, and refine.
The models address two primary concerns regarding code analysis: the complexity of vulnerability assessment and the financial implications of using AI for this purpose. Karbasi remarked on the challenges analysts face, stating, “That work is difficult because repositories are large, security signals are noisy, and the relevant evidence is rarely in one obvious place.”
Antares aims to mitigate these issues by providing a more cost-effective solution, particularly beneficial for public sector organizations, educational institutions, and smaller security teams. Karbasi highlighted that compact models lower inference costs and support local operations, allowing teams to maintain sensitive source code within their own environments.
