- Cisco is introducing Antares, a family of security small language models (SLMs) designed to identify known vulnerabilities within a codebase.
- The company is releasing two models, Antares-350M and Antares-1B, as open-weight models now available to the broader community.
- Benchmark testing shows that the Antares models outperform many powerful closed- and open-weight models in vulnerability localization at a fraction of the cost.
- The Antares models are compact enough to run locally, eliminating the need to send sensitive codebases to the cloud.
- Antares follows an iterative search pattern similar to how a human investigator works through a repository.
- The output of Antares includes a ranked list of source files likely to contain relevant vulnerabilities, along with the exploration trace that led to that result.
Cisco's Antares models, Antares-350M and Antares-1B, are designed to tackle the complex issue of identifying vulnerabilities in codebases. These models utilize an iterative search pattern similar to human investigation, producing a ranked list of source files likely to contain vulnerabilities.1256
The models are compact enough to run locally, allowing organizations to keep sensitive code secure without relying on cloud services. Cisco's benchmark testing indicates that Antares models outperform larger models, achieving 15.2 times lower costs compared to the best open-weight model, GLM-5.2, and 172 times cheaper than the strongest closed-source model, GPT-5.5.34
“Security can’t be a luxury good, yet advanced AI-based detection has largely belonged to organizations with frontier-scale budgets,” said a Cisco representative. “Antares’s results change that equation: near-frontier accuracy on secure code reasoning at a fraction of the cost.”
The Antares family is part of Cisco's broader initiative to enhance AI in security, following previous releases like Project CodeGuard and SecArena. The models are particularly aimed at universities, public-sector bodies, and smaller security teams that may lack the resources for more expensive solutions.
Cisco emphasizes that while Antares accelerates the initial vulnerability triage process, it does not replace comprehensive application security measures, including human review and dynamic testing.
“The Antares models are small enough to run on local hardware, letting security teams keep sensitive codebases in-house instead of sending them to the cloud. On Cisco's Vulnerability Localization Benchmark, Antares-1B outperformed larger models like GLM-5.2 and Gemini 3 Pro while costing under $1 per evaluation, versus $141 for GPT-5.5.”
SiliconANGLE
Help Net Security
