---
id: 6619
title: "We burned 11.7bn tokens to find the best cyber AI model"
source_domain: aikido.dev
source_url: "https://www.aikido.dev/blog/ai-model-benchmarks-aug-21-2026"
published_at: "2026-08-21T00:00:00Z"
tags: ["ai-security", "open-source", "benchmark", "aikido", "llm-security", "vulnerability-discovery", "ai-models", "deepseek"]
phying_url: "https://phying.com/item/6619"
lang: en
---

# We burned 11.7bn tokens to find the best cyber AI model

## Summary

Aikido burned 11.7B tokens testing 10 AI models against 32 fresh vulns: DeepSeek V4 Pro pooled the most finds (28/32), and open models now rival closed frontier ones—at the cost of noisier leads.

## Why it matters

This coverage gives security teams current context for monitoring, validation, and remediation.

## Intelligence Metadata

- **Sources:** aikido.dev
- **Published:** 2026-08-21T00:00:00Z
- **Original:** https://www.aikido.dev/blog/ai-model-benchmarks-aug-21-2026
- **Phying News:** https://phying.com/item/6619