LinkedIn·Monday, 24 August 2026·1d ago

Claude Mythos is arguably the strongest cybersecurity model available today, but it still came second on our benchmark. We tested the new…

Aikido Security

41,558 followers

Claude Mythos is arguably the strongest cybersecurity model available today, but it still came second on our benchmark. We tested the new Claude Security with Mythos and Codex Security with GPT-5.6 Sol against our own Code Security Audit by running them all on the same repo with 89 vulnerabilities. The results: Mythos found 60 of them for $157. GPT-5.6 Sol found 58 of them for $125. Aikido Code Security Audit found 68 of them for $75. Aikido’s Code Security Audit found eight more vulnerabilities, including 1 critical and 3 high-severity issues, than Mythos. More bugs found at less than half the price. Vulnerability discovery is a search problem. Claude Security runs its search on one expensive model, so every lead it chases costs Mythos money. Code Security Audit uses smaller specialized models to cover a lot of ground, and the expensive ones only get used for the hardest parts. Mythos missed a cross-tenant auth flaw that let users read other orgs' data, and a timestamp mismatch between the app and the database that kept expired API keys working. Code Security Audit, part of Aikido's AI Code Analysis, caught both. Frontier models are powerful, but as we’ve been saying for months now, the harness can affect the outcome even more than your model. Read the full story in Debarshi and Philippe Dourassov's post here: https://lnkd.in/eWgF44D4