Back to feed
News Story
BStandard68
Google AI (X)
2 sources

Google Launches Gemini 3.5 Flash Cyber AI Model for Cybersecurity

Google has launched Gemini 3.5 Flash Cyber, a specialized AI model for detecting and fixing cybersecurity vulnerabilities, built on Gemini 3.5 Flash and integrated into CodeMender. The model achieves competitive performance on benchmarks like CyberGym and is optimized for scale and lower cost. Due to dual-use concerns, it will initially be available only to governments and trusted partners via a limited-access pilot program.

SynthePulse Insight · AI deep reading

Google Launches Low-Cost AI Security Model Gemini 3.5 Flash Cyber, Challenging Anthropic Mythos

Version 1 · 1 source

Google releases a lightweight AI security model based on Gemini 3.5 Flash, designed for rapid vulnerability discovery and patching at a fraction of the cost of Anthropic's Mythos, outperforming similar models in V8 engine vulnerability detection.

  • Google launches Gemini 3.5 Flash Cyber, positioned as a low-cost alternative to large AI security models like Anthropic Mythos.
  • The model is integrated into the CodeMender secure coding agent, enabling high-speed, low-cost repeated calls to scan more code paths.
  • On the CyberGym benchmark, after up to five calls, performance is comparable to significantly larger models.
  • Discovered 55 unique confirmed issues in the V8 JavaScript engine, surpassing Gemini 3.5 Flash (47) and Opus 4.6 (36), with 10 issues found exclusively by this model.
  • Anthropic Mythos 5 is computationally intensive and costly, with usage costs twice that of Claude Opus 4.8.
  • The model is initially available to government and trusted partners through CodeMender.
Open section navigationProduct Launch and Positioning

Product Launch and Positioning

Google released Gemini 3.5 Flash Cyber on July 21, 2026, an AI security model designed for rapid vulnerability discovery and patching. According to Google's official blog, the model is described as a "cost-efficient and capable" alternative targeting larger, more expensive AI systems like Anthropic's Mythos.

Gemini 3.5 Flash Cyber is built on Gemini 3.5 Flash and is initially available through CodeMender, Google's secure coding agent, to government and trusted partners. CodeMender can call the model multiple times at high speed and low cost, enabling scanning of more code paths and vulnerability discovery.

Performance and Benchmarking

Google claims that on the CyberGym AI cybersecurity benchmark, Gemini 3.5 Flash Cyber achieved "competitive performance with significantly larger models" after up to five calls. Specifically, the model found 55 "unique confirmed issues" in the V8 JavaScript engine, compared to 47 for Gemini 3.5 Flash and 36 for Opus 4.6. Notably, 3.5 Flash Cyber discovered 10 issues that no other model found.

Google emphasizes that the model consistently discovers new code paths and vulnerabilities after multiple calls, suggesting its iterative scanning capability is key to its performance advantage.

Competitive Landscape: Anthropic Mythos

Anthropic's Mythos 5 is a powerful AI security model released under Project Glasswing, but it is computationally intensive and costly to use—reportedly twice the cost of Claude Opus 4.8. Microsoft has adopted Mythos for security checks and achieved its largest "Patch Tuesday" this month.

Google's release indicates a divergence in the AI security model market: one end features high-performance, high-cost models like Mythos, while Google focuses on low-cost, efficient alternatives. Additionally, China's Z.ai claims its model can compete with Mythos, but this claim has not been independently verified.

Credibility boundary

This article is primarily based on The Verge's coverage of Google's blog post. Performance data (e.g., V8 vulnerability counts) comes from Google's official statements and has not been independently verified by third parties. Cost and performance data for Anthropic Mythos also come from reports and may be based on Anthropic's public information.

Insight takeaway

Google enters the AI security market with a low-cost strategy; Gemini 3.5 Flash Cyber demonstrates competitive performance against larger models in specific benchmarks, but real-world deployment effectiveness and long-term reliability remain to be seen.

Sources for this version

  1. Google launches a cheaper alternative to large AI security models like MythosThe Verge AI

Primary report

Google AI (X)

Primary source

Same-event coverage

Also covered by 1 sources

The Verge AIT3Google launches cheaper AI security model Gemini 3.5 Flash Cyber