Google touts cheaper enterprise model

Alphabet’s Google used its I/O conference on May 19 to make two related arguments about artificial intelligence: its models are getting cheaper to run, and they need stronger defenses against manipulation. At the event in Mountain View, California, the company put AI agents into Search, expanded coding and task-automation tools, and introduced Gemini 3.5 Flash, which Reuters reported as a faster, lower-cost model aimed at business users. Google executives paired that product push with repeated claims about efficiency and enterprise spending. Separately, Google security and DeepMind teams have been publishing more detail about prompt-injection defenses as attacks on AI systems move from lab exercises toward real-world monitoring. ### Why did Google spend I/O talking about price? Sundar Pichai told reporters before the conference that heavy users could save more than $1 billion a year by switching workloads to Google’s models, according to Reuters. He said Google could deliver similar performance to other frontier models at up to one-third of the cost, framing price as a direct selling point for companies already watching token budgets. (insidetelecom.com) Google also cut the price of its top-tier AI Ultra subscription to $200 a month from $250 and added a $100-a-month version aimed at developers and work-focused users, Reuters reported. That pricing push came as Google tried to blunt competition from OpenAI and Anthropic in coding tools and enterprise deployments. ### What exactly is the cheaper model Google is pushing? (insidetelecom.com) Google launched Gemini 3.5 Flash on May 19 as a model designed for coding and automated tasks, Reuters reported, while saying Gemini 3.5 Pro would arrive next month. On Google Cloud’s pricing page, Gemini 3.5 Flash is listed at $1.50 per 1 million input tokens globally and $9 per 1 million output tokens. (insidetelecom.com) Google has also been building out lower-cost options below that tier. In a March 3 post, the company said Gemini 3.1 Flash-Lite would be offered to enterprises through Vertex AI at $0.25 per 1 million input tokens and $1.50 per 1 million output tokens, describing it as its most cost-effective Gemini 3-series model. ### How does that fit with Google’s broader enterprise pitch? (insidetelecom.com) Mountain View was the setting for a broader effort to show that Google can turn consumer scale into business revenue. Reuters said Google added agents that can make purchases, monitor ticket availability and plan schedules in real time, while also updating Antigravity, its coding assistant that competes with Anthropic’s Claude Code. (blog.google) Google Cloud has been making a similar case for weeks. In April, the company said nearly 75% of Google Cloud customers were using its AI products and that 330 customers had processed more than 1 trillion tokens each over the prior 12 months. Google also rebranded Vertex AI as Gemini Enterprise Agent Platform, pitching it as a place to build, govern and optimize agents for enterprise workflows. (insidetelecom.com) ### Why is Google talking more about attacks on AI outputs? Google’s security teams said in an April post that indirect prompt injection is now a top priority because attackers can plant hidden instructions in webpages, emails or documents that an AI system later reads. In that scenario, the model can follow the attacker’s instructions instead of the user’s request. Google said it had begun a broad sweep of the public web to monitor known injection patterns. (blog.google) Google DeepMind said in a separate paper that it has been running continuous adversarial evaluations against Gemini to test how the model behaves under adaptive prompt-injection attacks. The paper described those tests as part of an ongoing effort to make current and future Gemini versions more resilient in security-critical settings. ### What does that mean for customers buying these systems? (blog.google) Google’s own enterprise materials now put security and control next to model performance. The Gemini Enterprise Agent Platform page says customers can build, scale, govern and optimize agents, while Workspace updates this month introduced an AI control center for managing model and agent access to company data. (arxiv.org) June is the next concrete milestone Google has named publicly. Reuters reported that Gemini 3.5 Pro is due next month, and Google Cloud’s published pricing and enterprise control pages remain the main places where customers can track the rollout of models, grounding charges and admin tools. (insidetelecom.com) (cloud.google.com)