Cisco launches super affordable security language model called Antares (www.neowin.net)

Cisco has launched Antares, a new family of affordable small language models (SLMs) aimed at enhancing code security by identifying vulnerabilities. The Antares family includes three models—Antares-350M, Antares-1B, and Antares-3B—each named for their parameter size, with the two smaller models already available as open-weight versions on Hugging Face. Remarkably, these models outperform established systems like Gemini and GPT, offering effective vulnerability detection at a significantly reduced cost. Their small footprint allows companies to run them locally, ensuring confidential code remains secure. The significance of Antares lies in its innovative approach to vulnerability detection, blending traditional static analysis with advanced machine learning techniques. The models employ an iterative search strategy that replicates human investigative methods, starting with a vulnerability description and systematically narrowing down relevant code patterns. This capability supports various workflows, including locating files related to specific Common Weakness Enumeration (CWE) categories, enhancing static analyses, and facilitating local security assessments in compliance-heavy environments. By providing a cost-efficient, efficient tool for pinpointing vulnerabilities, Antares positions itself as a valuable asset in the ongoing battle against software security threats.