Google DeepMind Expands Gemini Flash Lineup with 3.6 Flash and Security-Focused 3.5 Variant

Google DeepMind expanded its Gemini Flash lineup on Monday, adding Gemini 3.6 Flash and, according to contemporaneous reporting by 9to5Google, two Gemini 3.5 variants: Flash-Lite and a restricted-access Flash Cyber model focused on security work.

The move matters because Flash is Google’s lower-cost, higher-speed model tier for high-volume AI tasks such as coding help, agent-style workflows and long-document processing. With this update, Google is doing more than refreshing a general-purpose model. It is also carving out a cyber-specific offering for vulnerability discovery and repair, a sign that model makers are increasingly segmenting products for mainstream application workloads and more sensitive security use cases.

On DeepMind’s live model page, Gemini 3.6 Flash is listed as a preview. Google says the model supports 1 million input tokens — a measure of how much text, code or other data can be fed into the model at once — and 64,000 output tokens. DeepMind’s pricing page lists Gemini 3.6 Flash at $1.50 per 1 million input tokens and $7.50 per 1 million output tokens.

DeepMind is also pitching efficiency as a key selling point. On the model page, the company describes Gemini 3.6 Flash as “Our workhorse model that reduces output token usage by 17% compared to 3.5 Flash, according to Artificial Analysis Index.” That matters in Flash’s target market because lower token use can reduce costs for developers running high-volume applications.

Google said 3.6 Flash is available across its broader Gemini ecosystem, including consumer, developer and enterprise products such as the Gemini app, the Gemini API and related Google tools.

The two 3.5 variants announced alongside it appear aimed at more specialized needs. According to 9to5Google’s report on the July 21 launch, Gemini 3.5 Flash-Lite is intended for high-throughput, low-latency work, positioning it as a cheaper option for applications where speed and scale matter more than top-end capability. Google has not provided pricing details in the materials cited here.

The more unusual addition is Gemini 3.5 Flash Cyber. 9to5Google reported that the model is designed for “finding and fixing security vulnerabilities,” putting it squarely in the defensive cybersecurity category. That kind of work sits close to dual-use risk because the same systems that help identify software weaknesses can also be misused.

Google is handling that rollout more cautiously. According to 9to5Google, the company’s CodeMender tool uses multiple 3.5 Flash Cyber agents, and access is initially being offered as a limited pilot for governments and trusted partners. In comments quoted by the publication, Google framed the project as a defensive measure: “This will give frontline defenders a head start in finding and fixing critical vulnerabilities before they can be exploited, while mitigating against broader misuse.”

The expansion builds on Google’s broader Gemini 3.5 push. On May 19, 2026, the company said it was introducing the Gemini 3.5 family, describing it as “our latest family of models combining frontier intelligence with action.” At that time, 3.5 Flash became the default model in the Gemini app and Google Search’s AI Mode, underscoring Flash’s role as the company’s practical, scaled deployment tier.

DeepMind has also published safety assessments for the Flash family. In the model card for Gemini 3.5 Flash, the company said additional testing found the model “remains below the cyber CCL” in its internal framework, a brief indication of the governance work now accompanying even lower-cost, high-volume AI releases.