Google announced Gemini 3.5 Flash on Tuesday. This is a new AI model that the company claims is the most powerful yet for coding and autonomous AI agents. The model, announced at the company’s annual Google I/O developer conference, allows developers to run coding pipelines independently, manage research projects, and build operating systems completely from scratch for internal testing.
This release marks a shift for Google from marketing AI as a conversational tool to marketing AI as an agent tool. It’s not just about answering questions; it’s about planning, building, and iterating on the actual work with minimal human input.
“3.5 Flash offers an incredible combination of quality and low latency, outperforming our latest frontier model, 3.1 Pro, on nearly every benchmark, including coding, agent tasks, and multimodal inference,” DeepMind chief engineer Koray Kavukcuoglu told reporters Monday ahead of general availability.
He added that it’s four times faster than other Frontier models, making it ideal for coding and agent tasks, but Google has “taken it to another level” by developing an optimized version of Flash that has the same quality but is 12 times faster.
According to Kavukcuoglu, this speed is central to Flash’s agent work design, where multiple AI agents run simultaneously on long-running tasks. On stage at I/O, Google engineer Varun Mohan demonstrated how agents are spawned to work in separate components before coming together to build a complete operating system within Antigravity, the company’s agent development platform and IDE.
Kavukcuoglu said Flash 3.5 was co-developed with Antigravity to give agents “a native environment in which they can live, work and run.”
At I/O, Google released Antigravity 2.0, a standalone desktop application designed around agent-first development.
The results are visible beyond the demonstration. Google says 3.5 Flash’s agent capabilities are already making an impact among its partners, including banks and fintech companies automating multi-week workflows, and data science teams gaining insights in complex data environments.
The model can run autonomously for several hours, but Tulsi Doshi, Google’s senior director and head of product, said it will occasionally pause and ask for user input if it encounters decision points or permission issues that require human judgment.
The two are designed to work together when Google releases its next 3.5 Pro model.
Doshi told TechCrunch, “3.5 Pro can be an orchestrator, a planner, and can actually leverage Flash to be different subagents. I think at the end of the day, it comes down to where do you really need that inference ability, do you really need a larger model that can really drive the inference side, or are there tasks where the ability to use a good brute force tool is really worth it?”
3.5 Flash is currently the default model for AI mode in Gemini apps and global search. At I/O, Google also announced agents features coming to Search that will allow users to create, customize, and manage AI agents directly on the platform. The new model also powers Gemini Spark, Google’s new personal AI agent designed to run 24/7 to help consumers manage their digital lives.
Delivering that level of AI functionality to the average consumer requires scrutiny. Google is currently facing a lawsuit after a man died by suicide last year after chatting with a Gemini for several weeks and then nearly causing a mass casualty incident.
The impact of harm would be even greater if powerful autonomous agents were made more widely available. Google says Gemini 3.5 includes enhanced cyber and CBRN (chemical, biological, radiological, and nuclear) security measures and is tailored to respond to sensitive questions rather than rejecting them outright.
Gemini 3.5 Flash is now generally available via Antigravity, Gemini API, Gemini Enterprise, and through AI mode in the Gemini app and Search.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
Source link
