Google's Gemini 3.5 Flash delivers 12x faster agentic AI

Google unveiled Gemini 3.5 Flash at its annual I/O developer conference, positioning the model as a cornerstone for agentic AI. According to DeepMind's chief technologist Koray Kavukcuoglu, the model offers an 'incredible combination of quality and low latency,' outperforming the previous frontier model 3.1 Pro across coding, agentic tasks, and multimodal reasoning benchmarks. A key differentiator is speed: Flash runs 4x faster than other frontier models, with an optimized version achieving 12x speed improvements at the same quality level. This speed is critical for autonomous agents that need to execute long-running, multi-step tasks concurrently.

The model was co-developed with Google's Antigravity platform, an IDE and development environment designed for agent-first workflows. At I/O, engineers demonstrated agents spawning sub-agents to build a full operating system independently. Beyond demos, early partners like banks and fintechs are using Flash to automate multi-week workflows. The model can run autonomously for hours, pausing only for human input at decision or permission points. Safety safeguards have been strengthened for cyber and CBRN risks. Gemini 3.5 Flash is now the default model in the Gemini app and AI Mode in Search, and will power Gemini Spark, a 24/7 personal AI agent. The forthcoming 3.5 Pro model will act as an orchestrator, delegating sub-tasks to Flash.