Security model Mythos achieves record 69% recall in defensive benchmark

jfiance · x · 2026-09-02

The closely watched cybersecurity model Mythos was evaluated on dfbench for open-ended defensive security work. Mythos achieved a 69% detection recall (the highest measured) and 24.5% precision on validation tasks. While highly capable, there remains a gap in precision compared to other frontier systems.

Related event: Security Model Mythos Hits Record 69% Recall but at High Cost, Low Precision(2 posts)→

Original post →

More from Models

Models channel →