GPT-6 Astra Tops Vercel DeepsecBench for Cybersecurity, 50% Cheaper Than Claude Opus 5

LauraModiano · x · 2026-09-06

rauchg shares that GPT-6 Astra is now the best cybersecurity model on Vercel's DeepsecBench, which measures how well models find security vulnerabilities in code, combining recall and precision.

Astra scored 37.79, finishing in 49 minutes what took GPT-5.6 Sol about 4 hours at similar cost. Against Claude Opus 5 max (32.44, $127.93, 2h34m), Astra is more effective and 50% cheaper at $63.70, with only 2 false positives and 97.9% precision.

Related event: GPT-6 Astra Tops Vercel's Security Benchmark at Half the Cost(2 posts)→

Original post →

More from Models

Models channel →