Cisco Antares AI models hunt bugs faster than Gemini and GPT, cost under $1 to scan 500 repos

Cisco's open-weight bug busters take on Google and OpenAI

Who needs expensive frontier models to find software vulns? Cisco has just released two open-weight models that specialize in finding known bugs in existing codebases. The models, Antares-350M and Antares-1B, are part of Cisco's new Antares family of security small language models (SLMs), and are now available on Hugging Face - but only to vetted users. "We're making sure we're gating that and appropriately granting access," DJ Sampath, Cisco's senior vice president and general manager of AI software and platform, told The Register. The company is working with academic and nonprofit organizations, as well as smaller and public organizations' security teams, to ensure they have access to the vulnerability-hunting models. Plus, because both are small models designed to run locally, "you also need the keys to the source code" to scan for and find vulnerabilities, Sampath said. "This means an attacker is going to be able to exploit an endpoint or a service that you have." It also means that proprietary code never leaves the organization's machines, compared to cloud-based LLMs that send code to the AI providers' external servers for processing and analysis. This enables security analysis in environments with strict privacy or compliance requirements, according to the networking and security giant. And yes, it's named after the massive red super-giant star. "It's almost 1,000 times bigger than the sun, even though the sun dominates the sky, and that is the analogy that we're using here for vulnerability detection and localization," Cisco VP and chief AI scientist Amin Karbasi told The Register. "The impact of vulnerabilities in your codebase is huge, but it might be only a single file or a few lines of code in a million lines of code." A future, 3-billion-parameter model in the Antares family won't be released to the public, Karbasi added. "We are completely gating the 3B model to make sure that we responsibly release it to communities that need it," he said. Small yet mighty Cisco claims that its models perform as well as or better than dozens of larger models in its new benchmark test that measures how efficiently AI models identify security flaws in codebases. Antares-1B outperforms Google's Gemini 3 Pro and is comparable to Z.ai's GLM-5.2, we're told, while the yet-to-be-released Antares-3B does a better job at finding vulnerabilities than GLM-5.2 and OpenAI's GPT-5.5. Plus, we're told that the small models scan code much faster and at a fraction of the cost of larger, token-gobbling AI systems. "If you look at the performance, in terms of the time it takes to finish 500 repositories, Antares finishes the entire cohort of repositories in 15 minutes, whereas frontier models take five hours," Karbasi said, adding that this translates to significantly less cost. "It takes like less than $1 whereas frontier models are above $100 into $150 of cost," he added. The difference, Karbasi explained, is that Cisco took a "fundamentally different approach" to building Antares. "These models have been trained in a very different way," he said. "Antares is inherently not a chatbot. It is an investigator. It is a search engine. It has to find a very specific thing that might be a needle in a haystack, and it goes and finds that." This required training the model on several different ways to search for vulnerabilities "because one way of search may not actually be fruitful, then it has to change its strategy, do it another way, and then do it another way," Karbasi said. "Because it is very nimble and it's very small, it can actually do a lot of search at the same time, which is very different from bigger models." Karbasi likened it to a bicycle on a busy London street: "You can go much faster than the biggest truck." Or, to use Sampath's favorite analogy for the benefits of using a small, security-focused model to find bugs in code: "Sometimes you don't need a private jet to go to a corner store, right?"®

Cisco's AI shrinks haystacks to help security teams hunt needles

Cisco is betting that enterprises will see value in having AI quickly identify the handful of files worthy of investigation by human software vulnerability researchers. Cisco has revealed a family of open-weight AI models called Antares that, it said, can help security teams isolate potentially vulnerable parts of a software repository before deeper investigation begins. Rather than detecting a specific CVE or generating a patch, these models search a codebase using only a Common Weakness Enumeration (CWE) description and return the files most likely to contain that class of vulnerability. "Its purpose is to reduce a large codebase to a focused set of files that a security professional or a downstream security workflow should investigate," Cisco's AI researcher Supriti Vijay said via email. "The goal is not to replace a security engineer's judgement or send them on a wild-goose chase, but to reduce fatigue and workload by helping them triage an issue earlier and focus their investigation on the most relevant parts of the codebase."

Cisco's tiny open-weight AI hunts bugs, beating Gemini

The networking giant's Antares models run on your own machines and, it claims, localise vulnerabilities faster and cheaper than frontier systems. It is gating access, and withholding its strongest model, because a bug-finder is also a bug-exploiter. Cisco's new Antares models are small enough to run on your own machines. And, it claims, they beat Google's Gemini and rival GPT-class systems at finding bugs. It is releasing them open-weight, though it is vetting who gets access. The pitch from Cisco is a quiet rebuke to the frontier-model arms race. You do not need a giant, pricey AI to hunt for software bugs. And you should not have to ship your source code to OpenAI or Google to do it. Cisco's Foundation AI team has released two open-weight security models, Antares-350M and Antares-1B, built to find known vulnerabilities buried in existing codebases. They sit on Hugging Face for the community, though Cisco is approving access case by case. It is holding a larger 3-billion-parameter model back from a full public release. These are not chatbots. Antares works a repository like a human investigator. It starts from a vulnerability description, searches the code, and reads candidate files. It backtracks when a path leads nowhere, then returns a ranked list of the files most likely to hide a flaw. The name comes from the red supergiant star, roughly 1,000 times the size of the sun. Small yet mighty Cisco says the little models punch far above their weight. On its own 500-task benchmark for vulnerability localisation, Antares-1B beats Google's Gemini 3 Pro and matches Z.ai's GLM-5.2. The unreleased 3B model, it claims, tops both GLM-5.2 and OpenAI's GPT-5.5. The real advantage shows in speed and cost. Chief AI scientist Amin Karbasi told The Register that Antares clears 500 code repositories in 15 minutes, against five hours for frontier models. That works out at less than $1, he said, versus $100 to $150 for the big systems. The trick is that Cisco trained the models to search, not to chat. Because each one is tiny and nimble, it can run many searches at once and switch tactics when a line of attack goes nowhere. Karbasi likened it to a bicycle weaving past a truck on a busy London street. Keeping the code at home Running locally is the other selling point. Being small and open-weight, the models keep proprietary code on a company's own machines. Cloud-based systems, by contrast, send that code to an AI provider's servers for scanning. For firms in regulated industries, keeping it in-house is the whole point. The catch There is a reason Cisco is holding back. A tool that finds flaws is also one that helps attackers exploit them. As Reza Shokri, a computer scientist at the National University of Singapore, put it, AI agents now "write more of the code, and are growing capable of exploiting it." So Cisco gates access to academic, nonprofit, and smaller public-sector security teams, and keeps its strongest model back. The aim, it says, is to stop advanced AI defence becoming a luxury good, just as attackers start wielding AI of their own. Cisco's bet is that the edge is no longer the size of the model. It is who gets to aim it.

Introducing Antares: Highly Efficient Open Weight AI Models for Vulnerability Localization

Collaborator(s): Supriti Vijay, Aman Priyanshu, Didier Chapoteau, Arthur Goldblatt, Kimia Majd, Fraser Burch, Jianliang He, Baturay Saglam, Takahiro Matsumoto, Zhuoran Yang Today, Cisco is introducing Antares, a family of security small language models (SLMs) purpose-built for one of the hardest, most time-consuming and expensive problems in security: pinpointing where known vulnerabilities exist within a codebase. We are releasing two of these models -- Antares-350M and Antares-1B -- as open-weight models now available to the broader community on Hugging Face. Benchmark testing shows that these models outperform many powerful closed- and open-weight models in this critical security task at a fraction of the cost. And they're compact enough to run locally, heading off the need to send sensitive codebases to the cloud. By combining open-weight models with Antares, open specifications with Foundry Security Spec, secure coding guidance with CodeGuard, and a new benchmark, we are working to define the types of practical, trustworthy AI tools that will help cybersecurity professionals do their job better, as well as cultivating an ecosystem and standards for further enterprise adoption of AI security tools. Figure 1. The Antares family of small language model outperforms a dozen larger models tested in the new Vulnerability Localization Benchmark. (Antares-3B is coming soon.) Small Models for a Costly and Sensitive Security Workflow Software security depends on connecting external vulnerability knowledge like public databases, advisories, and CWEs to internal code. That work is difficult because repositories are large, security signals are noisy, and the relevant evidence is rarely in one obvious place. Analysts often need to search through unfamiliar code, follow naming conventions, inspect call paths, compare candidate files, and decide whether a weakness is actually present. Additionally, organizations need security tooling that is not only capable, but practical to deploy. Compact models reduce inference costs, support local or on-premises operations, and help teams keep sensitive source code within their own environment. Antares can unlock the power of AI-assisted security for universities, public sector institutions, and smaller security teams that may have lacked the resources to use token-intensive AI models for critical tasks. Antares is purpose-built to address both problems. "Software has always shipped with vulnerabilities, and vast bodies of legacy code still carry latent flaws. AI now pushes on the problem from both sides: agents write more of the code, and are growing capable of exploiting it. That makes cheap, continuous verification a control that has to sit close to the loop, checking code before it lands. Small models are especially compelling here: they run locally, so proprietary code never leaves the machine, and they're fast enough to gate an agent's output in real time. Antares-1b is a good example: it recognizes vulnerabilities in code it is shown, across several weakness classes and languages. The provided CLI packages the model's capabilities into a practical workflow, making it easy to scan codebases, inspect findings, and integrate it into automated pipelines. What I found compelling was seeing a model this small navigate a codebase and identify security issues that would otherwise require larger models or more manual analysis." -- Reza Shokri, Associate Professor of Computer Science at the National University of Singapore (NUS) "Security can't be a luxury good, yet advanced AI-based detection has largely belonged to organizations with frontier-scale budgets. Antares's results change that equation: near-frontier accuracy on secure code reasoning at a fraction of the cost, fast enough to run on every commit. At a moment when attackers are already using AI, that kind of efficiency isn't a nice-to-have; it's what makes always-on security scanning possible for every team." -- Amin Saberi, Professor of Management Science and Engineering and Director of Language, Data, and Reasoning Lab at Stanford University What Makes Antares Different Traditional static analysis tools are valuable, but they are often rule-heavy and can produce results that require substantial triage. General-purpose coding models can reason about code, but they may not be optimized for security investigation, terminal navigation, or structured vulnerability localization. Antares is designed for that middle ground. Antares was inspired by pioneering research by the Cisco Foundation AI team showing that compact models can learn to search, reflect, revise their strategy, and backtrack when a path is unproductive. In other words, useful retrieval behavior can come from learned search strategies, not only from model scale. The next question was whether the same idea could help with an urgent real-world security problem: vulnerable code localization. Antares follows an iterative search pattern that resembles how a human investigator works through a repository. Each model starts from a vulnerability description, searches for relevant code patterns, reads candidate files, incorporates new evidence, changes direction when a path is unhelpful, and narrows toward the files most likely to matter. The goal is not to replace expert judgment, but rather to help make the first stages of source-code vulnerability triage faster, more repeatable, and easier to review. In practice, Antares can support workflows such as: * locating files related to a CWE category in a repository snapshot * triaging advisory-driven security investigations * augmenting static analysis with model-driven repository exploration * supporting CI/CD triage workflows where likely vulnerable files need early review * enabling local security analysis in environments with strict privacy or compliance requirements Antares outputs a ranked list of source files likely to contain a relevant vulnerability, along with the terminal exploration trace that led to that result. Security teams can use that output to prioritize analyst review during advisory triage, focused CWE investigations, or CI/CD security checks. Antares is not meant to replace the broader application security toolchain. Security teams still need dependency and software composition analysis, secret scanning, dynamic testing, infrastructure and container checks, threat modeling, remediation workflows, and expert review. Why We Needed a New Benchmark Building a model for vulnerable code localization raised an immediate measurement problem. We could not rely only on general coding benchmarks, because Antares does not target general issue resolution, patch generation, or broad code search. It is tailored for security-specific localization tasks: given limited security context, it identifies source files most likely to contain a vulnerability. The closest adjacent work is CodeScout, which evaluates terminal-based code-search agents on software-engineering benchmarks such as SWE-Bench Verified, Pro, and Lite. CodeScout shows that code localization can be trained and measured with a standard Unix terminal. But these benchmarks typically measure whether an agent can find code relevant to a software issue or development task, not whether it can localize vulnerable files from CWE-style security descriptions, security advisories, or vulnerability classes. This gap inspired the Vulnerability Localization Benchmark, a 500-task benchmark requiring models to simultaneously navigate unfamiliar codebases efficiently and recognize vulnerability patterns associated with specific CWE categories. Why Open-Weight We are releasing Antares as open-weight because the security community needs more accessible building blocks for practical, repository-level defense. Figure 2. On the vulnerability localization benchmark, the Antares model family completes runs at substantially lower estimated cost and runtime than larger comparison models. Vulnerability triage is expensive. It requires expertise, time, infrastructure, and access to sensitive code. Those costs are felt most sharply by organizations that already operate with constrained security resources: universities, research institutions, nonprofits, and public-sector teams that still maintain important software. A Focused Step Forward Antares is part of a broader Cisco effort to advance how security organizations use AI responsibly, measurably, and in service of defenders. With Foundry Security Spec, Cisco shared a model-agnostic blueprint for building agentic security evaluation systems with clear roles, guardrails, and reviewable outputs. With CodeGuard, Cisco helped contribute secure-by-default rules and skills that can guide AI coding agents toward safer software development. With Antares, we are adding compact security models and a benchmark focused on repository-level vulnerability localization. Cisco's efforts are connected by a common belief: AI in security has to move beyond impressive one-off demos and toward systems that practitioners can evaluate, govern, and improve. Foundry Security Spec helps define the harness around agentic security work. CodeGuard helps raise the floor for prevention by capturing secure coding practices in a reusable rules corpus. Antares helps with the expensive localization step, narrowing vulnerability intelligence into source files that human reviewers can inspect. Taken together, this work can help the community build a more durable foundation for AI-assisted defense: open specifications, reusable security knowledge, compact deployable models, and benchmarks that make progress measurable. The goal is to build toward a system where all security practitioners, regardless of on-prem or resource constraints, can effectively incorporate AI in everyday security operations. Visit our Hugging Face page to explore the Antares models and review the model card. To learn more about the benchmark and evaluation methodology, read the technical paper, or contact Cisco Foundation AI.

Cisco bets on small AI for cybersecurity

Why it matters: The models could give companies a cheaper way to repeatedly search large codebases for vulnerabilities without sending sensitive source code to an outside AI provider. Driving the news: Cisco debuted a new family of small language models, called Antares, that are specifically designed to investigate software repositories and identify where known vulnerabilities exist in a codebase, Amin Karbasi, Cisco's vice president and chief AI scientist, told Axios. * Cisco says the models performed similarly to much larger models, including OpenAI's GPT-5.5 and Z.ai's GLM-5.2, on Cisco's new benchmark that measures how accurately models identify vulnerable files in real-world software repositories. * Antares-350M and Antares-1B will be available Tuesday on Hugging Face to users Cisco verifies as cyber defenders. * Cisco is holding back the public release of a more capable 3-billion-parameter version, Antares-3B, which it plans to integrate into its own security products. Between the lines: Running large AI models across an organization's codebase can become expensive because defenders may need to rescan repositories as software changes. * In Cisco's testing, Antares scanned 500 repositories in roughly 15 minutes for less than $1, compared with roughly five hours and more than $100 for GPT-5.5. * The smaller models are also designed to run locally, allowing organizations to keep sensitive source code inside their own environments instead of sending it to third-party AI providers. * "You really don't need a private jet to go to your corner store," DJ Sampath, Cisco's senior vice president and general manager of AI software and platform, told Axios. "You want to be able to use something that's practical." Zoom in: Cisco trained Antares to behave more like a security investigator than a coding assistant, Karbasi said. * Rather than simply answering questions about code, the models learn how to search a repository, inspect files, change direction when a lead doesn't pan out and narrow in on the files most likely to contain a vulnerability. * "A vulnerability can be only a few lines of code in a million lines of code," Karbasi said. Antares "is going to be finding these security needles in the haystack of your repository." * General-purpose language models, by contrast, learn security tasks largely as an extension of broader coding capabilities. The intrigue: Cisco worked with U.S. government agencies on the models' safety and release, Sampath told Axios. * The company will also vet who can download the models to ensure adversaries and cybercriminals aren't gaining access to the tools. The big picture: Cisco is just the latest company to open-source tools designed to find and fix software vulnerabilities. Last week, Capital One open-sourced VulnHunter, an agentic AI security tool that takes an attackers' mindset and reviews source code for the most likely ways a hacker could break in. What to watch: Cisco is exploring an industry consortium to expand its work on open AI security tools, Sampath said.

Cisco releases Antares, open-weight small models for locating code vulnerabilities

Cisco Systems Inc. today introduced Antares, a family of small language models built to pinpoint where known security vulnerabilities sit inside a codebase, and released the first two as open-weight downloads on Hugging Face. The models come from Cisco Foundation AI, the company's research and engineering group focused on security-specific artificial intelligence. Antares targets vulnerability localization, the work of connecting external vulnerability data such as public databases, advisories and Common Weakness Enumeration entries to the specific files in a repository where a flaw is likely to live. That step is one of the slower and more expensive parts of software security. Analysts have to search unfamiliar code, follow naming conventions, trace call paths and compare candidate files before deciding whether a weakness is actually present. Antares is designed to narrow that search. Two models are available now, Antares-350M and Antares-1B, with a larger Antares-3B model due to follow. All are small enough to run locally, which lets security teams keep sensitive source code inside their own environment rather than sending it to a cloud service. Cisco is pitching that as a fit for universities, public-sector bodies, nonprofits and smaller security teams that lack the budget for token-heavy commercial models. Rather than scan a repository with fixed rules, Antares works through it the way a human investigator would. Each model starts from a vulnerability description, searches for relevant code patterns, reads candidate files, folds in new evidence, changes direction when a path leads nowhere and narrows toward the files most likely to matter. The output is a ranked list of source files likely to contain the relevant vulnerability, along with the terminal exploration trace that produced it. To measure the models, Cisco built its own test, the Vulnerability Localization Benchmark, a 500-entry set that requires a model to navigate an unfamiliar codebase and recognize vulnerability patterns tied to specific weakness categories. Existing code-search benchmarks measure whether an agent can find code relevant to a general software issue, the company said, not whether it can locate vulnerable files from security descriptions. The closest comparable work, a system called CodeScout, showed that code localization can be trained and measured with a standard Unix terminal. On that benchmark, Cisco said the Antares models beat about a dozen larger open- and closed-weight models on accuracy while running far cheaper. The company put the cost of an Antares evaluation at under $1, against roughly $12.50 for the strongest open-weight model it tested, GLM-5.2, and about $141 for the strongest closed-source model, GPT-5.5, figures it described as 15.2 times and 172 times cheaper. Antares also finished benchmark runs in about an hour, compared with roughly four and a half hours for GPT-5.5. Antares is the latest piece of a broader Cisco push around AI in security. In October, the company released Project CodeGuard, an open-source framework of secure-coding rules for AI coding agents and in May it published Foundry Security Spec, an open blueprint for building agentic security evaluation systems. Cisco Foundation AI has also shipped a run of open-weight security models over the past year, including Foundation-sec-8B and a reasoning version. "The goal is to build toward a system where all security practitioners, regardless of on-prem or resource constraints, can effectively incorporate AI in everyday security operations," Amin Karbasi, vice president and chief scientist at Cisco Foundation AI, wrote in a blog post. Cisco was clear about what Antares does not do. The models are meant to speed up the first pass of vulnerability triage, not replace it, and they do not stand in for the rest of the application security stack, including dependency and software composition analysis, secret scanning, dynamic testing and human review. The two available models and their model card are on Hugging Face, with the Antares-3B model still to come.