Google bets AI future on agents with Gemini 3.5 Flash

Google has officially launched Gemini 3.5 Flash, a new AI model unveiled Tuesday at the annual Google I/O developer conference. The company touts this as its most powerful model yet for coding and autonomous AI agents. In internal demonstrations, the model independently executed entire coding pipelines, managed research projects, and even built an operating system from scratch.

This release marks a clear strategic pivot for Google. Instead of positioning AI primarily as a conversational tool, the company is now pushing it as an agentic tool one that plans, builds, and iterates on real-world tasks with minimal human intervention.

Koray Kavukcuoglu, DeepMind’s chief technologist, briefed reporters Monday ahead of the public launch. “3.5 Flash offers an incredible combination of quality and low latency,” he said. “It outperforms our latest frontier model, 3.1 Pro, on nearly all the benchmarks,” including coding, agentic tasks, and multimodal reasoning. He noted that the model is 4x faster than other frontier models, a speed critical for coding and agentic workloads. Google has also developed an optimized version of Flash that is 12x faster while maintaining the same quality.

That speed is essential for Flash’s design, according to Kavukcuoglu. In agentic workflows, multiple AI agents run concurrently on long-running tasks. During the I/O keynote, Google engineer Varun Mohan demonstrated agents spawning to work on separate components before merging their outputs to build a complete operating system inside Antigravity, Google’s agentic development platform and IDE.

Kavukcuoglu explained that Flash 3.5 was co-developed with Antigravity to give agents a “native environment where they can live, work, and execute.” At I/O, Google also released Antigravity 2.0, a standalone desktop application built around agent-first development.

The impact is already tangible beyond staged demos. Google reports that 3.5 Flash’s agentic capabilities are delivering results with partners such as banks and fintechs automating multi-week workflows, and data science teams uncovering insights in complex data environments. The model can run autonomously for several hours. However, Tulsee Doshi, Google’s senior director and head of product, said it will occasionally pause to request user input when it encounters a decision point or permission issue requiring human judgment.

When Google releases its upcoming 3.5 Pro model, the two are designed to work together. Doshi told TechCrunch that “3.5 Pro becomes your orchestrator, your planner, and then it actually can leverage Flash to be the various sub-agents. I think it really comes down to where do you really want that reasoning power, where you actually want that larger model that can really push on the reasoning side versus where do you have tasks that really do merit good brute force tool use capabilities?”

Starting 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 agentic features 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 personal AI agent designed to run 24/7, helping consumers manage their digital lives.

Offering that level of AI capability to average consumers invites scrutiny. Google is currently facing a lawsuit after a man nearly committed a mass casualty event and died by suicide following weeks of chatting with Gemini last year. The potential for harm only grows when powerful autonomous agents become more widely available. Google says Gemini 3.5 has strengthened safeguards against cyber threats and CBRN (chemical, biological, radiological, and nuclear) risks, and is better calibrated to engage with sensitive questions rather than refusing them outright.

Gemini 3.5 Flash is available generally today through Antigravity, the Gemini API, Gemini Enterprise, the Gemini app, and AI mode in Search.