Cisco’s Latest Innovations: Antares Introduced
Who requires costly frontier models to discover software vulnerabilities? Cisco has just launched two clever models crafted to identify known issues in current codebases. Introducing Antares-350M and Antares-1B, the fresh faces in Cisco’s Antares line of security small language models (SLMs). They are now available on Hugging Face, but note that access is limited to approved users. “We’re ensuring that access is restricted and granted appropriately,” DJ Sampath, Cisco’s head of AI software, informed GadgetLad.
Staying Local
Cisco is collaborating with academic institutions and nonprofit organizations, along with some public security teams, to ensure they can utilize these vulnerability-detection models. As they are small models intended to operate on your premises, “you also need the keys to the source code” for vulnerability hunting, Sampath clarified. “This means an attacker could exploit a service or endpoint you possess.” However, rest assured, your valuable code remains on-site, unlike cloud-based LLMs that send your code to AI provider servers for processing.
The Powerful Antares Clarified
Named after the enormous red supergiant star, Antares is certainly not insignificant. “It’s nearly 1,000 times larger than the sun, even though the sun oversees the sky, serving as the analogy we are using here for vulnerability detection and localization,” Amin Karbasi, Cisco’s VP and chief AI scientist, shared with GadgetLad. “The ramifications of vulnerabilities in your codebase are significant, but it could just be a single file or a few lines of code out of a million.” There is a forthcoming 3-billion-parameter model in the Antares lineup, but don’t get too excited—Karbasi stated it’s secured more than a Geordie’s wallet.
Compact Yet Powerful
Cisco believes its new models achieve comparable performance, if not superior, to numerous larger ones in their newly established benchmark assessment. Antares-1B is putting up a strong challenge to Google’s Gemini 3 Pro and holds its own against Z.ai’s GLM-5.2, while the anticipated Antares-3B can identify vulnerabilities even more effectively than GLM-5.2 and OpenAI’s GPT-5.5. And here’s a surprise: these compact models process code more swiftly and economically than bigger, resource-heavy AI systems.
The Antares Advantage
Cisco asserts that Antares completes the entire set of 500 repositories in only 15 minutes, whereas the so-called frontier models drag on for five hours. That’s a substantial time and cost saving. “It costs less than $1, while frontier models range from $100 to $150,” Karbasi bragged.
A Bicycle in a Crowded Street
Cisco has taken a “fundamentally different approach” with Antares. This isn’t your typical chatbot—it functions as an investigator, a search engine tasked with locating a needle in a haystack. “These models have been trained in a very distinct manner,” Karbasi noted. “Antares is not fundamentally a chatbot. It operates as an investigator. It is a search engine. It has to pinpoint a very specific item that might be a needle in a haystack, and it successfully finds that.” It’s like a bicycle weaving through a congested London street: “You can move much quicker than the largest truck,” Karbasi remarked.
A Jet vs. a Quick Trip to the Local Store
Sampath encapsulates it well: “Sometimes you don’t need a private jet to visit a corner shop, right?” Just what you require when searching for an efficient, security-oriented model to detect bugs in your code. Here’s to a straightforward solution!
Overview: Vulnerability Hunting Without Overspending
There you have it, everyone. Cisco’s new Antares models are like a Jack Russell among a group of lazy hounds—small, swift, and perfectly effective. Who says you need a massive model when these little ones accomplish the task faster and more affordably? It’s time to let those bugs run! Check it out on gadgetlad.co.uk.
