Google has introduced three new Gemini models focused on faster AI agents, lower operating costs and software security.
Released on July 21, Gemini 3.6 Flash and Gemini 3.5 Flash-Lite are now available for developers and general users. Gemini 3.5 Flash Cyber will initially be restricted to governments and trusted partners because of its ability to identify and patch security vulnerabilities.
Gemini 3.6 Flash Reduces AI Agent Costs
Gemini 3.6 Flash is designed for coding, knowledge work, multimodal analysis and multi-step AI agents.
Google says the model uses 17% fewer output tokens than Gemini 3.5 Flash. Pricing is set at $1.50 per million input tokens and $7.50 per million output tokens, reducing the cost of running agents at scale.
Coding performance also improved. Gemini 3.6 Flash scored 49% on the DeepSWE benchmark, compared with 37% for the previous model, while making fewer unnecessary edits and using shorter execution loops.
The release arrives as Google continues expanding the physical infrastructure supporting its AI products, including a planned $1.5 billion expansion of its Alabama AI data centre.
Flash-Lite Targets High-Volume Tasks
Gemini 3.5 Flash-Lite is built for workloads where speed and cost matter more than advanced reasoning.
The model processes approximately 350 output tokens per second and costs $0.30 per million input tokens and $2.50 per million output tokens. Its main uses include document processing, data extraction, agentic search and automated subagent tasks.
Flash-Lite is available through the Gemini API, Google AI Studio, Android Studio and the Gemini app. It is also being introduced into Google Search.
Faster and cheaper models could accelerate the adoption of autonomous agents across finance and other industries. However, recent warnings about agentic AI and financial-market risks show that efficiency must be matched by stronger monitoring and safeguards.
Flash Cyber Focuses on Software Vulnerabilities
Gemini 3.5 Flash Cyber is a specialised security model designed to find, validate and patch vulnerabilities.
The model operates inside CodeMender, where several agents can examine different areas of a codebase before producing one combined security report. Google says it achieved competitive results on the CyberGym security benchmark.
Because vulnerability-detection technology can also be misused, Flash Cyber will not receive an immediate public release. Access will begin through a limited pilot for governments and trusted partners.
The restricted launch reflects the wider challenge of deploying powerful AI systems without creating new security risks. Similar concerns are influencing how businesses introduce AI agents into financial and blockchain infrastructure.
What Comes Next for Gemini
Gemini 3.6 Flash and Flash-Lite are available now, while Flash Cyber will enter its limited-access pilot separately.
Gemini 3.5 Pro remains in partner testing, with no confirmed public release date. Google has also started training Gemini 4 through what it describes as its most ambitious pre-training run so far.
The latest releases show Google prioritising practical AI deployment: lower token costs, faster agent workflows and specialised security tools. The next test will be whether developers can turn those improvements into reliable production systems without increasing operational and cybersecurity risks.
