Google Introduces Gemini 3.6 Flash, 3.5 Flash-Lite and a Cybersecurity-Focused Flash Model

Google announced three new Gemini Flash models, expanding its efficiency-focused AI lineup with Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and Gemini 3.5 Flash Cyber.

Three Flash models aimed at efficient AI workloads

In a post on Google’s official blog, the company described the new models as designed for lower-latency and lower-cost agentic AI workflows. The announcement positions the Flash family around practical deployment needs, including coding, multimodal use cases and cybersecurity tasks.

Axios also reported that Google introduced the three Gemini Flash models with an emphasis on efficiency, corroborating the company’s framing that the releases are meant to offer cheaper model options for developers and organizations.

Gemini 3.6 Flash adds multimodal reasoning and long context

Google DeepMind’s model card for Gemini 3.6 Flash describes it as a natively multimodal reasoning model. According to the model card, Gemini 3.6 Flash supports a context window of up to 1 million tokens and output of up to 64,000 tokens.

The same Google DeepMind model card says Gemini 3.6 Flash is distributed through an API and includes benchmark results and pricing comparisons published in July 2026. Those details suggest Google is presenting the model not only as a research milestone, but as a developer-facing product intended for production use through its API channels.

Google’s broader announcement says Gemini 3.6 Flash includes improvements for coding and multimodal tasks, although the company’s public materials should be treated as the primary source for those performance claims unless independently benchmarked.

Flash-Lite targets lower-cost use cases

Google also introduced Gemini 3.5 Flash-Lite as part of the same release. In Google’s announcement, Flash-Lite is grouped with the other new Flash models as part of a push toward lower-cost and lower-latency AI workflows.

The name and positioning indicate that Google is continuing to segment the Gemini family for different cost and performance needs. Based on the company’s announcement, Gemini 3.5 Flash-Lite is meant for users who prioritize efficiency, though Google’s provided source excerpt does not specify detailed benchmarks or technical limits for the Lite model.

Gemini 3.5 Flash Cyber focuses on vulnerability work

The most specialized release is Gemini 3.5 Flash Cyber. Google DeepMind says the model is built on Gemini 3.5 Flash and fine-tuned for cybersecurity work, specifically to find, validate and patch software vulnerabilities.

According to Google DeepMind, Gemini 3.5 Flash Cyber is used through CodeMender and is intended for governments and trusted partners. That framing is notable because Google is not presenting the model as a general consumer chatbot. Instead, the company is positioning it as a controlled cybersecurity tool for vulnerability discovery and remediation.

Google DeepMind describes the Cyber model as lightweight, which aligns with the broader Flash branding around speed and efficiency. Axios likewise reported that Gemini 3.5 Flash Cyber is intended for vulnerability detection and patching.

Why it matters

The release shows Google continuing to broaden the Gemini portfolio beyond a single general-purpose model. Gemini 3.6 Flash appears aimed at developers who need multimodal reasoning and long context, Gemini 3.5 Flash-Lite targets cost-sensitive workloads, and Gemini 3.5 Flash Cyber applies the Flash approach to software security.

The cybersecurity model may draw particular attention because vulnerability detection and patching are high-stakes applications. Google DeepMind’s description emphasizes trusted partners and government use, which suggests a more limited rollout than typical public model releases.

For developers and enterprise users, the key question will be how these models perform in real deployments against existing Gemini models and competing systems. For now, the clearest source-backed takeaway is that Google is expanding the Flash line with models designed around efficiency, long-context multimodal reasoning and specialized cybersecurity support.