Cisco Launches Antares: Lightweight AI Models for Fast Vulnerability Detection
Key Takeaways
- ▸Cisco releases Antares-350M and Antares-1B as open-weight security models on Hugging Face, designed specifically for vulnerability localization in code
- ▸Antares models outperform many larger closed and open-weight models on security benchmarks while requiring significantly lower computational cost
- ▸The compact models support local deployment, enabling organizations to keep sensitive source code on-premises while integrating into automated security pipelines
Summary
Cisco has introduced Antares, a family of open-weight small language models (SLMs) purpose-built for vulnerability localization in code. The company is releasing two models—Antares-350M and Antares-1B—as open-weight models on Hugging Face, with a third model (Antares-3B) coming soon. According to benchmark testing, these compact models outperform many larger closed and open-weight alternatives on security tasks while requiring a fraction of the computational cost and can run locally to keep sensitive source code on-premises.
Antares addresses one of cybersecurity's most time-consuming and expensive challenges: efficiently pinpointing where known vulnerabilities exist within large codebases. By combining these models with complementary initiatives—the Foundry Security Spec, CodeGuard guidance, and a new Vulnerability Localization Benchmark—Cisco is establishing practical standards for enterprise AI security tools. The models are particularly valuable for universities, public sector institutions, and smaller security teams that previously lacked resources to deploy token-intensive AI models for critical security tasks.
- Initiative includes open specifications, coding guidance, and benchmarking to establish interoperable standards for enterprise AI security adoption
Editorial Opinion
Antares represents a meaningful democratization of AI-powered security. By delivering frontier-class vulnerability detection in models small enough to run locally and inexpensive enough for any organization, Cisco is making advanced code security accessible to institutions that previously couldn't afford such capabilities—a critical shift as AI systems themselves become both more prevalent and more capable of exploiting vulnerabilities. The emphasis on open-weight models and interoperable standards signals that security tooling, unlike many AI applications, must build trust through transparency rather than proprietary control.
