Google launches Gemini 3.5 Flash, eyes token cost, performance balance

Google launched Gemini 3.5 Flash, a series of models that aims to combine frontier intelligence, actions and reasonable token costs. The company also teased Gemini 3.5 Pro, rolled out Gemini Omni and Omni Flash, revamped its Antigravity development platform, upgraded AI Mode, redesigned the Gemini App and rolled out a set of personal and information agents.

Alphabet CEO Sundar Pichai kicked off the announcements at Google I/O, a developer conference that features a hefty dose of consumer features that will wind up in Google Cloud for enterprises. Simply put, you have to pay attention. Here's a look at what was announced and what will play in the enterprise.

Gemini Flash 3.5. Pichai said Gemini Flash 3.5 has revamped how Google works internally. "We focused on agent decoding, long horizon tasks and real-world workflows," said Pichai, who touted better benchmarks for economically valuable tasks. "Gemini 3.5 Flash is very capable model at the frontier but remarkably fast. Flash 3.5 delivers frontier level capabilities at less than half the price and in some cases a third of the price."

• Google Cloud Next 2026: A look at the big themes
• How Verizon manages AI agent sprawl
• Citigroup's wealth unit launches Citi Sky, a Google-powered wealth advisor
• The big picture behind Ask Macy’s, a Gemini powered AI agent
• Merck inks Google Cloud agentic AI deal worth up to $1 billion

Why it matters to the enterprise is clear: Companies are just now seeing the token bills for AI agents and the sticker shock is real. "You've heard the anecdotes from other CIOs that companies are already blowing their annual token budgets and it's only May," said Pichai. "If companies used a mix of Flash and other models they could save a lot of money."

Gemini 3.5 Flash is being rolled out across Google's products and APIs. Now that Gemini 3.5 Flash can support multi-agent autonomous sessions it can run complex coding pipelines, iterative research projects long-running projects.

In a nutshell, Gemini 3.5 Flash sounds great. Sounds greater if it can help you control your token budget. Here's what CxOs need to know.

Gemini 3.5 Pro and Gemini Omni. Pichai said that Gemini 3.5 Pro is rolling out next month and will be embedded into Antigravity 2.0. While Gemini 3.5 Pro will improve on Google's previous frontier models the way Pichai moved past it indicates that the game has changed. For enterprises, and Google since it has to service new features for its own business, it's about agentic architecture and completing tasks well with fewer tokens.

Gemini Omni and Omni Flash falls into a similar bucket. Omni is Google's new multimodal world model to generate video today and images and text later from multimodal inputs. Omni represents a new architecture relative to Veo and it's fun. Omni will also kill your token budget so enterprises are going to play with this as consumers. Developers and enterprises will get API access in the weeks ahead.

The upshot for Google's latest models is that they are trained on the company's latest TPUs (8T and 8I) with a distributed training architecture that can train larger models in weeks instead of months.

• Google Cloud presses full-stack AI edge with new TPUs, Agentic Data Cloud, Gemini Enterprise Agent Platform
• Google Cloud CEO Kurian on AI model evolution, CPUs importance and being open

Antigravity 2.0 and 2.4. One of the loose ends from Google Cloud Next was a lack of a cohesive agent-first development platform. Pichai said Antigravity has been reimagined and built on Gemini 3.5 Flash. Antigravity 2.4 is a standalone desktop app designed to orchestrate teams of agents.

In addition, Antigravity 2.0 now has a CLI terminal and SDK. The upshot is that Google has rallied behind Antigravity as its developer play and theoretically can better compete with Anthropic Claude Code/Cowork and OpenAI Codex. To be fair, Google only launched Antigravity six months ago.

Agents Payments Protocol (AP2) and Universal Commerce Protocol (UCP) expansion. Google is expanding UCP to more verticals including hotels and local food delivery and regions. AP2 will include privacy preserving tech and verifiable links between user, merchant and processors. Google also rolled out a new intelligent shopping cart that collects items from Google services, monitors price changes, deals and restocks, reasons and optimizes for payment card perks.

These additions will matter to any enterprise that has to play the agentic commerce game. The intelligent shopping cart will roll out across search and the Gemini app in the summer with YouTube and Gmail to follow.

Rest assured that commerce will play a huge role in Google's new global AI search experience that'll include a new conversational search box and agents that can monitor topics.

Gemini Spark, a 24/7 personal agent. Google outlined an always-on agent that can continue tasks when devices are off, integrates with Google services and can manage email workflows, study guides events, small business inboxes and other tasks. Gemini Spark rides along with a morning agent that runs overnight to brief you on what's in Gmail, Calendar, Docs, Sheets, Slides and Chrome.

According to Google, Gemini Spark will get multiple MCP connectors, custom sub-agents and use AP2 guardrails for payments.

There are two reasons Gemini Spark is worth monitoring. First, the economics. Gemini Spark is always on and it'll be interesting to see what Alphabet has to say about the engagement and cost tradeoffs.