The original title is: "Cisco's Antares Models Cost $1 to Find Security Bugs. GPT-5.5 Costs $141"

Short summary

Cisco released Antares-350M and Antares-1B, open-weight models that find security vulnerabilities by linking CVEs to specific files. On Cisco's 500-entry benchmark, Antares scanned repos in under $1 while GPT-5.5 cost $141 and took 4.5 hours. The article argues specialized small models that run locally will outcompete frontier models on cost and speed for narrow tasks like security scanning.

• •Cisco's Antares models scan codebases for known vulnerabilities for under $1 vs $141 for GPT-5.5
• •Models run locally, keeping sensitive source code on-premises
• •Specialized small models offer compounding efficiency advantages over frontier models for narrow tasks

Generated with AI, which can make mistakes.