Google launches Gemini 3.6 Flash and a restricted cybersecurity model

Google is expanding its Gemini model family with two lower-cost general-purpose models and a cybersecurity-focused system that is initially limited to governments and trusted partners. The releases underline Google’s effort to compete on the price, speed and operating efficiency of AI tools, particularly for companies building automated workflows.

The most capable of the new public models, Gemini 3.6 Flash, is designed for coding, document work and tasks involving text, images and data. Google says it uses 17% fewer output tokens than Gemini 3.5 Flash in the Artificial Analysis Index, while also charging less per token. Tokens are the units of text that AI systems process and generate, so lower token use can reduce the cost of recurring AI work.

Google prices Gemini 3.6 Flash at $1.50 per million input tokens and $7.50 per million output tokens. The company says the model takes fewer reasoning steps and tool calls in multi-step tasks. That could matter for AI agents that search documents, extract information, draft reports or complete actions across business software.

Focus on high-volume AI work

Google also releases Gemini 3.5 Flash-Lite, its fastest and cheapest model in the 3.5 series. It targets high-volume work such as document processing, agentic search, data extraction and smaller tasks assigned by larger AI systems. Google says the model can generate 350 output tokens per second, according to Artificial Analysis measurements.

Flash-Lite costs $0.30 per million input tokens and $2.50 per million output tokens. It includes configurable reasoning settings, allowing developers to choose faster and cheaper responses for simple tasks or more extensive processing for multi-step assignments.

The models are available through the Gemini API, Google AI Studio and Android Studio. Gemini 3.6 Flash is also available in Google’s enterprise agent platform and Gemini Enterprise app. Flash-Lite is rolling out in the Gemini app and Google Search.

Limited-access „Cyber“ model

Google’s third release takes a more cautious path. Gemini 3.5 Flash Cyber is a specialized model for detecting, validating and repairing software vulnerabilities. It works within CodeMender, Google’s code-security agent system, where multiple AI agents collaborate on a combined security report.

Google says Flash Cyber is fine-tuned from Gemini 3.5 Flash and delivers competitive results on the CyberGym benchmark at a lower price per token than larger models. Because tools that identify software weaknesses can also be misused, Google limits access to a pilot for governments and trusted partners. The company says this approach is intended to give defenders earlier access while reducing broader risks.

The wider rollout comes as AI companies face pressure to deliver capable models at a sustainable cost. In its report for CNBC, MacKenzie Sigalos notes that Google is positioning the cybersecurity model as an answer to Anthropic’s early lead in automated code defense. The report also highlights growing competition from Chinese model developers and the importance of having enough computing capacity to serve popular models.

Google says it is testing Gemini 3.5 Pro with partners and has begun pre-training Gemini 4. The company also says it continues to explore closer links between its models, custom chips and cloud infrastructure to reduce the cost of operating AI at scale.

Sources

• Google expands Gemini lineup with cheaper models and new Mythos rival – CNBC
• Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber – Google Blog