With Gemini 3.5 Flash, Google is betting the next AI wave on agents, not chatbots

Google on Tuesday launched Gemini 3.5 Flash, a new AI model that the company says is the most powerful yet for programming and autonomous AI agents. The model, which was presented at the company’s annual Google I/O developer conference, can independently execute lines of coding, manage research projects, and in internal testing, build an entire operating system from scratch.

The release marks Google’s shift from promoting AI as a conversational tool to AI as an agentic tool. It’s not just answering questions, but planning, building and iterating on real work with minimal human input.

“3.5 Flash provides an amazing combination of quality and low latency. It outperforms our latest frontier model, 3.1 Pro, on almost all benchmarks, including programming, agent tasks, and multimodal reasoning,” Koray Cavoglu, chief technology officer at DeepMind, told reporters on Monday ahead of the public launch.

He added that it is 4 times faster than other leading models, which is an ideal speed for programming and proxying tasks, but Google has “taken it to another level” by developing an improved version of Flash that is 12 times faster and with the same quality.

This speed is key to Flash’s design for agent work, where multiple AI agents work simultaneously on long-running tasks, according to Kavukcuoglu. On stage at the I/O conference, Google engineer Varun Mohan showed customers getting to work on separate components before coming together to build a full operating system inside Antigravity, the company’s proxy development platform and IDE.

Kavukcuoglu said Flash 3.5 was developed in conjunction with Antigravity so customers could have a “native environment where they can live, work and execute.”

At I/O, Google released Antigravity 2.0, a standalone desktop application designed around agent-first development.

The gains show up outside of the demos. Google says Flash Agent 3.5’s capabilities are already creating impact among partners, such as banks and fintech companies that have been automating workflows for weeks, or data science teams searching for insights into complex data environments.

The model can run autonomously for several hours, though Tulsi Doshi, senior director and head of product at Google, said it will occasionally stop and ask for user input when it reaches a decision point or permission issue that requires human judgment.

When Google launches its upcoming 3.5 Pro model, the two are designed to work side by side.

Doshi told TechCrunch that “the 3.5 Pro becomes your orchestrator, your planner, and then it can actually leverage Flash to be the different sub-agents. I think it really comes down to where do you really want that power of reasoning, where do you actually want that larger model that can really push on the logical side versus where do you have tasks that really deserve good brute force tool capabilities?”

Now, 3.5 Flash is the default model in the Gemini app and in AI mode in search globally. At I/O, Google also announced agent capabilities coming to search, allowing users to create, customize, and manage AI agents directly on the platform. The new model will also power Gemini Spark, Google’s new personal AI agent designed to work 24/7 to help consumers manage their digital lives.

Providing this level of AI capability to everyday consumers comes after scrutiny. Google is currently facing a lawsuit after a man nearly committed a mass shooting and died by suicide weeks after chatting with a Gemini last year.

Harmful effects only increase when powerful independent factors become more widely available. Google says Gemini 3.5 has enhanced cyber, biological, radiological and nuclear safeguards and is better calibrated to handle sensitive questions rather than rejecting them outright.

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