Cisco's Antares: Local Code Review at a Fraction of GPT's Cost

shashib · x · 2026-08-12

Amin Karbasi, head of Cisco Foundation AI (and former Yale professor), introduced the Antares project at the Black Hat conference.

Antares addresses the privacy痛点 of traditional code security tools that require sending code externally. It uses small models trained to search codebases like an investigator working a lead, running entirely within the customer's local environment so the code never leaves.

Efficiency metrics reveal striking cost savings: a 500-task sweep across 290 repositories costs only $0.82 with Antares, compared to $141 with GPT-4.

Related event: Cisco Open-Sources Antares for Cost-Effective Code Review(2 posts)→

Original post →

More from coding & agent

coding & agent channel →