Google's newest model is named Gemini 3.5, and the tagline explains what's changed: "frontier intelligence with action."
That word - action - is doing a lot of work. Google's announcement positions Gemini 3.5 as a step beyond a model that answers questions and into one that completes tasks. In AI research, this is called "agentic" behavior: the model can decide to search for information, act on what it finds, and continue working through a multi-step problem without someone manually pushing it along at each stage.
What "action" means in practice
A standard AI model gives you one response and waits. An agentic model - say, one you've asked to research a competitor and write a summary - searches the web, reads the results, identifies gaps, searches again, then drafts the summary. The model decides the sequence, not you. Previous Gemini versions could do some of this, but Google is signaling that 3.5 was specifically optimized for multi-step work rather than treating it as a secondary feature.
This is the same direction OpenAI has been pushing with ChatGPT's operator-style features and Anthropic has been building toward with Claude's computer use capabilities. The race at the frontier right now isn't about who scores highest on a reasoning benchmark - it's about which model can actually finish a job without constant hand-holding.
Google's Distribution Advantage
Gemini 3.5 will power AI Mode in Search, Gemini Advanced, and Google Workspace tools. That matters more than the model specs for most people. Almost everyone who uses Google Search is now a potential daily user of Gemini 3.5, whether they know it or not. OpenAI and Anthropic have to convince people to download an app or visit a new URL. Google's model ships inside the product two billion people already open every day.
For developers and businesses evaluating which AI platform to build on, the context window (how much text the model can process at once - critical for tasks like analyzing long documents), API pricing, and rate limits are what will determine practical adoption. Google hasn't published all those specifics yet, but given how competitive model pricing has become in the past year, those numbers will be heavily scrutinized when they land.
For daily AI tool users, the near-term change is simpler: the Google products you already use are about to get a meaningfully more capable model underneath them.