Google has launched Gemini 3.5 Flash, a new AI model designed to support agentic workflows across its products and enterprise platforms, as the company looks to move generative AI beyond chatbot-style interactions and deeper into business operations.

The model, announced at the annual Google I/O developer conference, is available through the Gemini app, AI Mode in Google Search, Google Antigravity, the Gemini API in Google AI Studio and Android Studio, Gemini Enterprise Agent Platform, and Gemini Enterprise.

In a blog post, Google said Gemini 3.5 Flash is built for tasks including software development, financial document preparation, customer onboarding, OCR, tax workflows, and data diagnostics.

Google also sought to position the model as a faster alternative to larger flagship systems. It described Gemini 3.5 Flash as its strongest model yet for agentic and coding tasks, claiming it outperforms Gemini 3.1 Pro on benchmarks including Terminal-Bench 2.1, GDPval-AA, and MCP Atlas.

The company also said the model leads in multimodal understanding, with a score of 84.2% on CharXiv Reasoning.

“When looking at output tokens per second, it is 4 times faster than other frontier models,” Google added.

Google also said it worked with industry partners to develop the Gemini 3.5 model series, adding that they “are seeing meaningful impact — from banks and fintechs automating multi-week workflows to data science teams unearthing insights amidst complex data environments.”

Gemini’s enterprise test

Analysts said Gemini 3.5 Flash should be seen less as an improved chatbot and more as part of Google’s push to build AI agents that can carry out enterprise tasks under supervision.

“Google’s speed, cost, and performance improvements matter because many AI pilots fail when they become too slow or expensive at scale,” said Pareekh Jain, CEO of Pareekh Consulting. “Faster and cheaper models could make AI agents practical for real business operations like coding, support, analytics, and automation.”

But CIOs should focus not only on model costs, but also on the cost of completing a workflow, such as resolving a claims exception, reviewing a contract, triaging a service incident, or moving a software fix through testing and approval, according to Sanchit Vir Gogia, chief analyst at Greyhound Research.

“Vendor benchmarks test capability. Enterprise pilots test survivability,” Gogia said.

Neil Shah, vice president for research at Counterpoint Research, said that enterprise goals are also changing. “The enterprise objective has been evolving from summarizing a document or answering prompt-based questions or basic code generation to deploying supervised, autonomous background workers directly into core business workflows,” Shah said.

That raises the question of whether Google can make agentic AI reliable enough for production use, not just faster or cheaper to run.

As AI agents move from passive assistants to active participants in business processes, enterprises will also need stronger controls over how they operate, said Anushree Verma, senior director analyst at Gartner.

“Enterprises face a new set of challenges as AI agents are adopted across business systems, for example, what actions are agents authorized to perform and under what circumstances,” Verma said.

The risks extend beyond operational errors. Agents operating across multiple systems could widen the attack surface, creating new entry points for attackers and increasing the chance that malicious prompts or data trigger unintended actions, she said.

“Accountability, auditability, explainability are going to be key concerns as well, and observability starts becoming very critical as you deploy more agents,” Verma added. “There are more issues that need to be handled with agents rapidly getting adopted, which can create an agent sprawl.”