Google has introduced three new Gemini AI models focused on lower-cost enterprise workloads, including a cybersecurity-specific model designed to automate vulnerability detection and secure code analysis.
Noticeably absent, however, was Gemini 3.6 Pro—the flagship model many developers have been expecting.
The releases reflect a broader shift in enterprise AI, where inference costs, latency, and workload-specific performance increasingly matter alongside benchmark scores. Rather than competing on raw capability alone, Google is emphasizing models businesses can deploy more efficiently.
Balance, speed, and efficiency define Google’s latest models
AI models have traditionally been judged by one measure above all else: raw intelligence.
But as enterprises move beyond experimentation and deploy AI in production, practical considerations such as response speed, token efficiency, operating costs, and the ability to handle a broad range of tasks have become just as important as benchmark performance.
That shift has driven demand for models that strike a balance between capability, efficiency, and scalability rather than excelling in only one area. It is against this backdrop that Google introduced Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, touting both as models that combine strong performance with lower latency and reduced operating costs for real-world enterprise workloads.
Gemini 3.6 Flash is Google’s latest general-purpose Flash model, built for coding, multimodal reasoning, agentic workflows, and instruction following while reducing output token usage. Gemini 3.5 Flash-Lite, meanwhile, targets high-volume, cost-sensitive workloads where speed and affordability often matter more than frontier-level reasoning.
Another cybersecurity model joins the growing queue
The last of the three models released on Tuesday does the opposite of Google’s broad-based approach of AI model releases. The company released Gemini 3.5 Flash Cyber, a low-cost cybersecurity-focused AI model.
According to Google, the model is “built on top of 3.5 Flash, and fine-tuned for finding and fixing cybersecurity vulnerabilities at a lower price per token than larger models.”
The company further noted that Gemini 3.5 Flash Cyber performed comparably with cybersecurity-focused models from Anthropic and OpenAI.
Google says Gemini 3.5 Flash Cyber performs comparably with cybersecurity-focused models from Anthropic and OpenAI while offering a lower price per token. If those claims hold up in production, the model could become a lower-cost option for organizations looking to automate security workflows.
However, unlike Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, Google says Gemini 3.5 Flash Cyber won’t be made immediately available for public use, citing “an intentional approach” against abuse.
Google’s flagship and next-generation models still underway
Google’s latest announcements stop short of its most anticipated AI release. The company said Gemini 3.6 Pro remains in testing after falling short of internal expectations in some coding tasks.
Beyond that, Google also confirmed Gemini 4 is in development as its next-generation frontier model. Google said Gemini 4 is under development as its next-generation frontier model, although it shared few technical details or release timing.
Google is following a broader industry shift
Google’s latest releases aren’t just about expanding the Gemini lineup — they’re a response to where the AI market is heading.
As enterprises move beyond pilots and embed AI more deeply, factors such as cost, speed, and the model balance are becoming just as important as raw model intelligence.
That shift is already reshaping competition. Chinese AI developers have gained attention by doing exactly this, putting pressure on rivals to prove value beyond benchmark scores. Google’s latest Flash models and the company’s plans to develop efficient ways to run AI models without watering down their capabilities suggest it sees the same trend.
For enterprises and professionals, this cross-border competition is good news as it creates cheaper, more efficient models across different regions — a relief for those who operate internationally. It also means that they have more choice of AI providers.
More News: Google is reportedly developing a new AI chip that could make Gemini up to 10 times more power-efficient by embedding parts of the model directly into silicon.
