Google I/O 2026: Gemini Spark personal AI agent launches with Gemini 3.5 Flash

The Agentic Gemini Era

Google opened Google I/O 2026 on May 19 at Shoreline Amphitheatre in Mountain View with two headline announcements: Gemini 3.5 Flash, its new default frontier model, and Gemini Spark, a persistent personal AI agent. Both fit under a single strategic framing: AI that acts, not just answers.

Gemini 3.5 Flash

Gemini 3.5 Flash tops Gemini 3.1 Pro on every benchmark Google ran, including Terminal-Bench 2.1 (76.2%), MCP Atlas (83.6%), and GDPval-AA (1656 Elo), while running 4× faster in output tokens per second than comparable frontier alternatives. API pricing is $0.50 per million input tokens and $3.00 per million output tokens—roughly half the cost of comparable frontier options. The context window is 1 million input tokens with a 64k output cap. It is now the default engine for the Gemini app, AI Mode in Google Search, and the Gemini API globally as of May 19. Gemini 3.5 Pro, the larger sibling already in internal use at Google, is scheduled for a public rollout in June 2026.

Gemini Spark

Gemini Spark is Google’s answer to the 24/7 personal agent category. Powered by Gemini 3.5 and the Google Antigravity agentic harness, Spark can run long-horizon tasks in the background—even while a device is off. Users can reach it via a dedicated Gmail address, and the agent can interact directly with the web through Chrome. Third-party tool integrations via MCP are coming in the following weeks.

The rollout starts with trusted testers, expanding to Google AI Ultra subscribers in the US the week of May 26. Planned summer features include authorized payments with budget controls, creating custom sub-agents, and text or email interaction with Spark.

Other I/O Highlights

AI Overviews has surpassed 2.5 billion monthly active users; AI Mode hit 1 billion monthly active users within its first year. Google previewed Android XR smart glasses. Creative platforms joining the Gemini app include Adobe Firefly AI Assistant, Canva Magic Layers (live for AI Ultra subscribers), and CapCut (coming soon). Gemini Omni, a new model series accepting image, audio, video, and text and outputting video, was also announced.

Source: Google Blog, TechCrunch