With Gemini 3.5 Flash, Google is betting its next wave of AI on assistants, not chatbots

Google launched on Tuesday Gemini 3.5 Flash, a new version of AI that the company says is more powerful in coding and autonomous AI assistants. The model, which was introduced at the company’s annual conference Google I/O, can install automated writing pipelines, manage research projects, and, in internal testing, create operating systems from scratch.

This release marks Google’s shift from positioning AI as a communication tool to AI as a support tool. It’s not just answering questions, but planning, building, and repeating real work with minimal human input.

Koray Kavukcuoglu, DeepMind’s chief technology officer, told reporters on Monday before the public launch that: “3.5 Flash offers amazing integration and low latency. It outperforms our latest version, 3.1 Pro, in almost all benchmarks,” including text, agetic tasks, and multitasking.

He said that it is 4x faster than other border types, a speed that is good for writing documents and agent work, but that Google has “taken it to another level” by creating an optimized version of Flash that is 12x faster than the same type.

That speed is critical to Flash’s production of workflows, where multiple AI assistants work for long periods of time, according to Kavukcuoglu. Onstage at I/O, Google engineer Varun Mohan, showed the collaborators starting to work on different components before coming together to create an operating system within Antigravity, the company’s development platform and IDE.

Kavukcuoglu said Flash 3.5 was developed together with Antigravity so that agents have “a place where they can live, work, and kill.”

At I/O, Google released Antigravity 2.0, a standalone app built around the first development.

The benefits are more obvious than the demos. Google says that the 3.5 Flash’s agetic capabilities are already generating interest among partners, such as banks and fintechs who are creating workflows of several weeks, or data science teams finding information in the most difficult environments.

The model can run autonomously for several hours, although Tulsee Doshi, Google’s chief product officer, said it would sometimes pause and ask for the user’s input when it comes to a decision or permission that requires human judgment.

When Google releases its upcoming version 3.5 Pro, the two are designed to work together.

Doshi told TechCrunch that “3.5 Pro becomes your caller, your scheduler, and then it can upgrade Flash to different agents.

Now, 3.5 Flash is the default version in Gemini software and in AI Mode in Search worldwide. At I/O, Google also announced assistant capabilities coming to Searchallowing users to create, edit, and manage AI agents on the platform. The new model will also be power Gemini SparkGoogle’s new AI assistant designed to work 24/7 to help consumers manage their digital lives.

Delivering the full range of AI to the average consumer comes with an assessment. Google is currently facing a lawsuit after a man nearly killed himself after a few weeks of dating Gemini last year.

The effects of injury are only exacerbated by making independent power tools more widely available. Google says Gemini 3.5 has strengthened cyber and CBRN (chemical, biological, radiological, and nuclear) security and is better managed to accommodate complex questions rather than rejecting them outright.

Gemini 3.5 Flash is available today through Antigravity, Gemini API, and Gemini Enterprise, as well as through the Gemini app and AI features in search.