Aikido beats Claude Security at vulnerability scanning with more findings at half the cost
cyb3rops · x · 2026-08-24
Aikido benchmarked its Code Security Audit against Anthropic's Claude Security (running Mythos) and Codex Security on the same target: a private benchmark app with 89 known vulnerabilities. Claude Security found 60/89 (67%) at $157, Aikido found 68/89 (76%) at $75, and Codex Security 58/89 at $125. Aikido found 8 more vulnerabilities than Claude Security, including 1 critical and 3 high severity, at less than half the cost. Key takeaway: vulnerability discovery is fundamentally a search problem — the harness around the model matters as much as model capability, and many small-model agents beat a few big-model agents.
More from coding & agent
- How to choose the right thinking level for your AI workflows — brandon_galang · 2026-08-25
- AgentSky Launches as OpenRouter for Agents with Unified API and Benchmarking — FellMentKE · 2026-08-25
- Idea: Autonomous agent loop to fix slow queries via automated PRs — mattpocockuk · 2026-08-25
- Developer builds Flue AI agent to diagnose and fix 'slop' in web design — craigsdennis · 2026-08-25
- Grok Bot Plugins in Action: Retrieval, Bi-directional Sync, and Cross-channel Workflows — mattyp · 2026-08-25
- Hermes agents form autonomous crew but spam emails on autopilot — NickPassig · 2026-08-24