Google launches Gemini 3.5 Flash for AI agents

Google has launched Gemini 3.5 Flash, an artificial intelligence model designed to handle complex tasks, write code and work for longer with little supervision. It is already available to users, developers and businesses across several Google products.

The company presents this version as the start of the Gemini 3.5 family. The model combines fast responses with the ability to plan, use tools and complete multistep processes, which Google calls agentic tasks.

What Gemini 3.5 Flash can do

An AI agent does more than answer a question. It can break a goal into steps, make intermediate decisions and act through tools. For example, it can analyze a codebase, suggest changes, test them and fix errors without the developer having to direct every move.

Gemini 3.5 Flash is designed specifically for this kind of work. Google says it is its most capable model yet for agents and coding, with applications including:

• Creating and maintaining applications.
• Analyzing large volumes of data.
• Preparing financial documents.
• Automating processes that could previously take weeks.
• Generating web interfaces and more interactive charts.

According to tests published by Google, Gemini 3.5 Flash outperforms Gemini 3.1 Pro on several benchmarks, which are standardized tests used to compare models. It scores 76.2% on Terminal-Bench 2.1, 1656 Elo on GDPval-AA and 83.6% on MCP Atlas. In multimodal understanding, which combines text, images and other formats, it reaches 84.2% on CharXiv Reasoning.

These figures come from Google's evaluation and do not guarantee the same performance on every real-world task. The company also says the model generates tokens, the units it uses to produce text, four times faster than other frontier models.

Available across Google products

Gemini 3.5 Flash is reaching billions of people through the Gemini app and Google's AI Mode in Search. It is also available to developers and businesses through:

• Google Antigravity, the agent-focused development platform.
• Gemini API in Google AI Studio and Android Studio.
• Gemini Enterprise Agent Platform.
• Gemini Enterprise.

Google is also making it the default model for the Gemini app and AI Mode in Search globally.

For you, the most visible change will be that some Gemini and Search features can move from explaining something to working on it. Instead of merely showing information about a topic, an agent can organize it, create a visual representation or complete a task based on your instructions.

A personal agent that works all day

Google has also introduced Gemini Spark, a personal agent based on 3.5 Flash. The proposal is for it to run 24 hours a day and help manage the user's digital life, always under their direction.

The rollout begins with a group of test users. Google plans to bring the beta to Google AI Ultra subscribers in the United States next week. For now, it is not a general feature available to all users.

The same model powers new Search features, including information agents that can work continuously and interactive visual experiences. The idea is for the search engine to do more than deliver links. It should also help investigate a subject and present the result in a more dynamic way.

Gemini 3.5 Pro is also in preparation

Google says it is already using Gemini 3.5 Pro internally and expects to launch it next month. It has not yet detailed its results or availability terms, so the announcement of that model should not be confused with a completed launch.

The company also highlights that Gemini 3.5 was developed under its safety framework for advanced models. According to Google, it includes new measures to reduce the generation of harmful content and avoid unnecessary refusals for safe requests, along with tools to analyze the system's internal reasoning before it responds.

The important part of this launch is not only that Gemini responds better, but that Google is positioning it as a tool capable of taking action. The next thing to watch will be whether these agents can reliably complete long tasks, properly understand the limits of what they are allowed to do and explain their decisions when they work without constant supervision.