Google launches Gemini 3.6 Flash for more efficient AI agents

Google has launched Gemini 3.6 Flash, describing it as a faster and more efficient model for developers building AI agents, coding tools and enterprise workflows while reducing the cost of deploying AI at scale.

According to Google, the model uses 17% fewer output tokens than Gemini 3.5 Flash on the Artificial Analysis Index and requires fewer reasoning steps and tool calls for multi-stage tasks, reducing the cost of running production AI agents.

Google has priced Gemini 3.6 Flash at US$1.50 per million input tokens and US$7.50 per million output tokens. The company says it improves coding, computer use, document analysis, chart interpretation and report drafting while reducing unnecessary edits and execution loops.

The company also reported gains on benchmarks including DeepSWE and OSWorld-Verified, alongside stronger safeguards against cyber-offence and chemical, biological, radiological and nuclear misuse.

Google also introduced Gemini 3.5 Flash-Lite, its fastest and lowest-cost model in the 3.5 family, targeting high-volume workloads such as search, document processing and data extraction.

In addition, a specialised Gemini 3.5 Flash Cyber model will power Google’s CodeMender security agent to detect, validate and patch software vulnerabilities. Owing to its potential for misuse, access will initially be restricted to governments and trusted partners through a pilot programme.

Gemini 3.6 Flash and Flash-Lite are available through the Gemini API, Google AI Studio, Android Studio, Gemini Enterprise and the Gemini app, with Flash-Lite also rolling out in Google Search.

Google said Gemini 3.5 Pro remains in partner testing while work is already underway on what it describes as its most ambitious pre-training run yet for Gemini 4.

Why does it matter?

The launch reflects a broader shift in AI development from simply increasing model size towards improving efficiency, reliability and cost-effectiveness for real-world deployment. As AI agents become more widely adopted, reducing inference costs and improving task execution are becoming increasingly important competitive advantages.

The restricted release of Gemini’s cybersecurity model also illustrates how AI developers are adopting more differentiated access policies for high-risk capabilities. Rather than making every model broadly available, companies are increasingly limiting advanced cyber tools to trusted users while attempting to balance defensive benefits with concerns about misuse.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!