Cisco Released Open-Source AI Models for Cybersecurity — Antares-350M and Antares-1B

A New Era in Cybersecurity: Small Open Models That Outperform Giants

In July 2026, Cisco released two new open-source AI models for cybersecurity — Antares-350M and Antares-1B. These models are designed to detect vulnerabilities in software code, and according to Cisco, they fundamentally change the economics of security analysis. Even the smallest model — Antares-350M — catches approximately 150 times more vulnerabilities per dollar than large AI agents like Cognition's Devin Security Swarm.

This announcement is particularly significant against the backdrop of the cybersecurity industry's search for more accessible and effective solutions. Cisco's approach demonstrates that sometimes smaller, specialized models deliver more value than their gigantic counterparts.

Benchmark Results: Antares vs. GPT-5.5

According to Cisco's tests, reported by Axios, Antares scanned 500 code repositories in about 15 minutes for under $1. The same task took GPT-5.5 five hours and cost over $100. This means Antares is roughly 100x faster and 100x cheaper than OpenAI's flagship model for the specific task of detecting cybersecurity vulnerabilities.

Cisco developer Aman Priyanshu stated on X that Antares-350M "catches about 150 times more vulnerabilities per dollar" compared to large AI agents. These figures are based on internal Cisco testing comparing Antares's performance against Cognition's Devin Security Swarm.

Technical Details and Architecture

Antares-350M (350 million parameters) and Antares-1B (1 billion parameters) are transformer-based models specifically trained for cybersecurity tasks. According to the technical report, the models were trained on approximately 72% security-concept data and 15% code search histories. This means the models understand not just code syntax but also security-specific nuances — SQL injections, XSS, buffer overflows, and other common vulnerabilities.

Both models are open-source and available on Hugging Face, enabling any company to use them on their own infrastructure. The models run locally, which is especially important for organizations handling sensitive code — code never leaves the company's boundaries.

Privacy and Local Deployment

One of the key advantages Cisco emphasizes is that Antares models run entirely locally. This means that a company's source code is never sent to external servers — all analysis happens within the same infrastructure where the code is stored.

This is particularly crucial for financial, healthcare, and defense industries where code confidentiality is critical. Many organizations cannot send their code to external AI services due to security risks. Antares solves this problem precisely through local execution.

Cisco's 3B Model: The Giant They Kept for Themselves

Interestingly, Cisco decided to keep the larger 3-billion-parameter version for its own products. According to the technical report, this version performs close to GPT-5.5 and beats open-source models up to 200 times its size.

This strategy resembles other technology companies' approaches — releasing smaller open-source models to the community while retaining more powerful versions for commercial products. In Cisco's case, the 3B model will likely be integrated into their existing security products such as Cisco SecureX and others.

Exploring an Industry Consortium for Open AI Security Tools

Cisco is also exploring the creation of an industry consortium for open AI security tools. This move signals that the company sees the importance of an open-source ecosystem in cybersecurity and wants to encourage collaboration with other industry players.

Such a consortium could be a significant step toward faster development and wider adoption of open AI security tools. If major companies join forces, it could create a security tool ecosystem accessible to everyone.

Why This Matters for DevSecOps

Antares models are important because they make DevSecOps practices more accessible. Today, many companies cannot implement regular code vulnerability scanning because traditional tools are either too expensive or too slow.

With Antares, integration into CI/CD pipelines is simple and affordable. The ability to scan 500 repositories in 15 minutes for under $1 means companies can check every commit without significant costs. This is especially important for small and medium businesses that lack dedicated security teams.

Implications and the Road Ahead

Cisco's release of the Antares models demonstrates that the future of AI in cybersecurity isn't necessarily large, expensive models — it's small, specialized, efficient solutions. This trend may soon spread to other domains — finance, healthcare, legal technology.

Open-source availability, local deployment, and under $1 cost for 500 repositories — all of this makes Antares one of the most interesting AI releases of 2026. Cisco's 3B model's reported performance, approaching GPT-5.5 levels, further reinforces the promise of this direction.

For the cybersecurity industry, this means that security analysis automation is becoming more accessible than ever. Antares may be the tool that brings open-source AI into the cybersecurity mainstream, enabling organizations of all sizes to protect their code without breaking the bank.