- Cisco has released Antares AI models specifically designed to identify code vulnerabilities.
- Antares models consume significantly fewer tokens than general-purpose models and can run locally in customer-controlled environments, protecting sensitive source code.
- In benchmark testing, Antares achieved a 500-entry evaluation in ~15 minutes on a single GPU for less than $1, making it 15x cheaper than the leading open model and 172x cheaper than the leading frontier model.
- Antares is aimed at public sector institutions and smaller security teams that may lack resources for token-intensive models, and it supports air-gapped and closed-network deployments.
- Cisco AI Researcher Supriti Vijay noted that frontier models, while capable, are not always the best for specialized security workflows due to cost and restrictions.
- Vijay emphasized that enterprise buyers should evaluate systems based on the specific task rather than model size or pressure to adopt frontier AI, and expect more agentic workflows with associated risks.
- Cisco's approach focuses on scaling capability through efficient, task-specific learning rather than relying on model size alone.
Cisco's Antares models, including Antares-350M and Antares-1B, are designed to enhance code vulnerability detection in software repositories while significantly reducing operational costs.2
The models are tailored for local deployment, ensuring that sensitive source code remains secure and private.
According to Cisco, running a full 500-entry evaluation on Antares takes about 15 minutes on a single GPU and costs less than $1, making it 15x cheaper than leading open models and 172x cheaper than frontier models.3
This cost efficiency is particularly beneficial for public sector institutions and smaller security teams that may lack the resources for token-intensive AI solutions.4
“With Antares, our focus was on scaling capability through efficient, task-specific learning rather than relying on model size alone,” said Vijay, a Cisco representative.8
He also noted that enterprise buyers should expect software security workflows to become more agentic in the coming years, potentially improving code analysis speed but also introducing risks such as excessive permissions and unintended tool actions.
“Buyers should evaluate systems based on the specific task they need to solve, rather than model size or pressure to adopt frontier AI because others are doing so,” Vijay advised.67
“The models consume significantly fewer tokens and can run locally inside customer-controlled environments to protect sensitive source code. In benchmark testing, a 500-entry evaluation cost less than $1 on a single GPU, making it 15x cheaper than the leading open model and 172x cheaper than the leading frontier model.”
CX Today
