Cisco Releases Antares: Efficient Open-Weight AI Models for Identifying Code Vulnerabilities
Key Takeaways
- ▸Cisco releases two open-weight small language models (Antares-350M and Antares-1B) optimized for finding code vulnerabilities in existing codebases
- ▸Models achieve near-frontier accuracy on vulnerability detection at a fraction of the cost of larger models, with performance benchmarked against a dozen competing systems
- ▸Compact design enables local, on-premises execution, keeping proprietary source code secure and avoiding cloud inference costs
Summary
Cisco has introduced Antares, a family of security small language models (SLMs) purpose-built for vulnerability localization in codebases. The company is releasing two models—Antares-350M and Antares-1B—as open-weight models on Hugging Face, with Antares-3B coming soon. Benchmark testing shows these compact models outperform many larger closed and open-weight models on vulnerability detection tasks while requiring significantly lower inference costs.
What sets Antares apart is its practical design for enterprise security workflows. The models are small enough to run locally on-premises, eliminating the need to send sensitive source code to cloud services—a critical concern for organizations handling proprietary applications. The release includes supporting infrastructure: open specifications via Foundry Security Spec, CodeGuard for secure coding guidance, and a new Vulnerability Localization Benchmark to measure performance.
The initiative addresses a significant gap in AI security tooling. While advanced AI-based vulnerability detection has largely been accessible only to large organizations with substantial budgets, Antares aims to democratize this capability for universities, public sector institutions, and smaller security teams. By packaging the models with practical CLI tools and integration capabilities, Cisco is working to establish standards and an ecosystem for enterprise AI security adoption.
- Release includes supporting tooling (CLI, specifications, benchmark) designed to establish standards for enterprise AI security workflows
Editorial Opinion
Antares represents a meaningful democratization of AI security capabilities—bringing frontier-level vulnerability detection within reach of organizations that can't afford massive cloud inference costs. However, the release highlights the inherent tension in open AI security models: the same tools that help defenders find weaknesses could lower barriers for adversaries. The real measure of success will be whether enterprises can deploy these responsibly and integrate them into secure development practices faster than attack sophistication evolves.
