Cisco has released two new open AI models, Antares-350M and Antares-1B, designed to detect vulnerabilities in software code. Developer Aman Priyanshu stated on X that the smallest version identifies roughly 150 times more security flaws per dollar than large agents like Cognition’s Devin Security Swarm. Internal testing by the company showed Antares scanned 500 code repositories in approximately 15 minutes for under one dollar, whereas GPT-5.5 required five hours and cost over $100 for the same task. Both models operate locally on company hardware, ensuring sensitive code never leaves the premises. The technical report notes the models were trained on roughly 72 percent security-concept data and 15 percent code search histories. Cisco retains a larger three-billion-parameter version for its own products, which reportedly matches GPT-5.5 performance while beating open models up to 200 times its size. The firm is also exploring an industry consortium for open AI security tools.
This approach prioritises efficiency and data privacy over the raw power of massive language models. Local execution prevents data leakage, a critical concern for enterprises handling proprietary source code. The shift suggests that specific security tasks do not always require the computational expense of generalist giants.
- Antares-350M and Antares-1B are fully open models.
- Scanning 500 repos costs under $1 and takes 15 minutes.
- A larger 3B parameter version matches GPT-5.5 performance.
