Security teams who can't send source code off-site get small, free, Apache-licensed models, built narrowly for locating known vulnerabilities.

Kenning's take

What it does

Antares is a family of small security language models from Cisco built to pinpoint where a known vulnerability exists in a code repository. Cisco released Antares-350M and Antares-1B as open-weight models on Hugging Face in July 2026, says a 3B version is coming, and reports that they beat larger closed and open models on its new Vulnerability Localization Benchmark. They are small enough to run on a local machine so source code does not leave the organisation, and come with a command-line tool for scanning codebases.

  • Antares-350M and Antares-1B open-weight models
  • Locates known vulnerabilities in code repositories
  • Runs locally so code stays on the machine
  • Command-line tool for scanning and pipelines
  • Apache 2.0 licence on Hugging Face

Who it suits. Security teams, universities and public-sector bodies that want low-cost, local AI help with vulnerability analysis.

Key facts

Category
AI models
Made by
Cisco (Foundation AI)
Released
21 July 2026
How to get it
free download of the weights from Hugging Face (fdtn-ai/antares-350m and fdtn-ai/antares-1b)
Works on
Hugging Face, local
In the UK
Yes
Free plan
fully free: open weights under Apache 2.0

FAQ

When did Cisco release Antares?

Cisco announced Antares and released the 350M and 1B models on 21 July 2026.

Is Cisco Antares free?

Yes. The models are open weight under the Apache 2.0 licence on Hugging Face.

What does Antares do?

It finds where known vulnerabilities exist within a codebase, and is small enough to run locally.

Facts from the maker's own pages and the sources listed. Checked 2026-10-11.