Claude Mythos 5.2 Arrives in Devin Security, Scores Rise in Vulnerability Tests

At a glance

Devin Security now offers Claude Mythos 5.2 to eligible enterprise customers. Security teams can use it without separate access to the models. The model scored much higher on Cognition's internal security benchmark.

Key points

1. 1

   Mythos 5.2 scored 73.5 on Cognition's internal security benchmark.

2. 2

   The previous version, Mythos 5.1, scored 56.9, and Claude Opus 5.5 scored 32.7.

3. 3

   The benchmark uses 25 repositories in several languages and frameworks, with 1,781 known public vulnerabilities.

4. 4

   Cognition said it will offer current and future enterprise customers free scans, backed by hundreds of millions of dollars in credits.

Why it matters

It helps companies gauge which AI models are better at finding security flaws as new threats emerge.