Cisco has launched two small, open AI models designed for cybersecurity tasks, focusing on vulnerability detection in software code. The models, named Antares-350M and Antares-1B, are marketed as cost-effective solutions for identifying security flaws. According to developer Aman Priyanshu, the smallest model detects about 150 times more vulnerabilities per dollar than large AI agents such as Cognition's Devin Security Swarm. Cisco's claim is based on performance metrics that highlight the models' efficiency and affordability in comparison to existing solutions.
In internal testing, Cisco reported that Antares-350M scanned 500 code repositories in approximately 15 minutes for under a dollar. In contrast, GPT-5.5 completed the same task in five hours and at a cost exceeding $100. The company emphasized that its models offer the best value for money in vulnerability detection, making them an attractive option for organizations seeking efficient cybersecurity tools. Both models are designed to run locally, ensuring that sensitive code remains within the company's network and does not leave the organization.
The technical report states that the models were trained on approximately 72 percent security-concept data and 15 percent code search histories. Cisco is retaining a larger three-billion-parameter version for its own products, which reportedly performs close to GPT-5.5 and outperforms open models up to 200 times its size. The company is also exploring the formation of an industry consortium to promote open AI security tools.
Source: thedecoder
