Google releases Gemini 3.6 Flash and a limited cyber-focused model for CodeMender

Google has introduced three Gemini models aimed at making production AI agents cheaper, faster, and more specialized. The July 21 announcement adds Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber, a security-oriented model that Google says will be paired with its CodeMender vulnerability-fixing agent.

The broadest release is Gemini 3.6 Flash, which Google describes as its new workhorse model for coding, knowledge work, and multimodal tasks. The company says the model uses 17% fewer output tokens than Gemini 3.5 Flash on the Artificial Analysis Index, and in some DeepSWE benchmark settings it observed reductions of up to 65%. Google also published pricing of $1.50 per million input tokens and $7.50 per million output tokens, positioning the release as a cost move as much as a quality update.

Why it matters

The launch is another sign that AI competition is moving from headline benchmark scores toward the operating costs of multi-step agents. Agentic systems can call tools, inspect files, run code, and loop through intermediate steps, so small changes in latency, verbosity, and token use can materially affect production bills. Google says 3.6 Flash takes fewer reasoning steps and tool calls for multi-step workflows, while improving scores in coding, computer use, and knowledge-work evaluations compared with 3.5 Flash.

Gemini 3.5 Flash-Lite is aimed at higher-volume workloads. Google calls it the fastest model in the 3.5 series and says it reaches 350 output tokens per second, with pricing of $0.30 per million input tokens and $2.50 per million output tokens. The company says it is suitable for agentic search, document processing, translation, summarization, and other tasks where throughput and cost are central constraints.

A narrower cyber release

The most restricted part of the announcement is Gemini 3.5 Flash Cyber. Google says the model is built on 3.5 Flash and fine-tuned to find and fix cybersecurity vulnerabilities at lower cost than larger models. Inside CodeMender, multiple Flash Cyber agents produce a combined report, and Google says the system reaches competitive frontier performance on CyberGym.

Because vulnerability discovery and patch generation can be dual-use, Google is not making the cyber model generally available. It says Gemini 3.5 Flash Cyber will be offered soon only to governments and trusted partners through a limited-access CodeMender pilot. By contrast, Gemini 3.6 Flash and 3.5 Flash-Lite are available now through the Gemini API in Google AI Studio and Android Studio, Gemini Enterprise, and the Gemini app, with Flash-Lite also rolling out in Google Search.

Sources

• blog.google

Cover photo by panumas nikhomkhai on Pexels, used under the Pexels License.