Google Releases Three New A.I. Models

By the Editorial Team · Wednesday, July 22, 2026

Google introduced three new AI models, including Gemini 3.6 Flash, its most advanced model, and Gemini 3.5 Flash Cyber, a specialized tool for cybersecurity designed to detect and fix software vulnerabilities efficiently. By limiting access to trusted partners and reducing operational costs, Google aims to enhance cybersecurity defenses while competing with rivals like OpenAI and Anthropic in the rapidly evolving AI landscape.

Google unveiled three new artificial intelligence models on Tuesday, including Gemini 3.6 Flash, its most advanced offering to date, and Gemini 3.5 Flash Cyber, a model specifically fine-tuned for cybersecurity applications. The company aims to strengthen its position against competitors such as Anthropic and OpenAI, both of which have made significant advances in AI-driven cybersecurity tools this year.

Gemini 3.6 Flash represents an improvement over Google’s earlier Flash version, which was highly rated for coding and financial tasks in various AI benchmarks. The updated model is also designed to be more cost-efficient for users, utilizing 17 percent fewer tokens, thereby reducing operational expenses. The second model, Gemini 3.5 Flash Cyber, focuses on identifying and repairing software vulnerabilities more affordably than larger AI systems, according to Google. The third model, Gemini 3.5 Flash-Lite, is tailored for managing AI "agents"—autonomous bots that function as digital personal assistants.

“These Flash models are built to balance efficiency and quality,” said Tulsee Doshi, senior director of product management on Google’s Gemini team.

Google is also developing Gemini 3.5 Pro, a flagship model expected to deliver the highest performance among its AI suite. Although initially slated for release in June, the Pro model remains in testing. Google has stated it will make the model available “as soon as it’s ready.”

Google’s intensified investment in AI follows the surge ignited by OpenAI’s ChatGPT chatbot in 2022. While Google Gemini models have ranked among the top AI tools, the company has recently faced competition from new entrants including Anthropic, OpenAI, and Chinese startups such as Moonshot AI. This month, Meta introduced a new model that reportedly surpasses Gemini’s capabilities on several leaderboards.

Cybersecurity has emerged as a critical focus area for AI. Such models can rapidly detect and address software flaws, enhancing defensive capabilities. However, they also carry risks if misused for cyberattacks. Anthropic, for instance, released the cybersecurity-oriented model Mythos in April, restricting its access to a limited group of organizations to safeguard against malicious exploitation. The company subsequently rolled out Fable, designed with guardrails to prevent hacking misuse. OpenAI followed a similar path, initially offering its cybersecurity model to select organizations for defense preparation before a broader release.

Google plans to restrict access to Gemini 3.5 Flash Cyber to governments and trusted partners. In internal testing, the model identified 55 issues in the V8 JavaScript Engine, an extensively used open-source codebase, including 10 vulnerabilities that previous models had not detected. Doshi emphasized that this approach aims to give defenders an edge in patching critical vulnerabilities before they can be exploited, while limiting the risk of malicious use.

All three new Flash models will be offered at lower costs compared to Google’s prior AI offerings, reflecting improvements in both performance and efficiency.

This article was reported by synthesizing coverage from multiple independent news outlets to present a balanced, unbiased account. Read about our editorial standards. Spotted an error? Let us know.