Google I/O 2026: Google announces Gemini 3.5 models and Gemini Spark AI agent

At the Google I/O 2026, the Mountain View tech giant unveils new flagship AI models, the Gemini 3.5 Flash and Gemini 3.5 Pro models. The company claims that the new generation Flash model is four times faster than other frontier models. 

Alongside the new AI models, Google also launched Gemini Spark, a new personal AI agent that can remotely run 24/7 even when the device is off. Gemini Spark can connect with Google apps such as Sheets and Gmail to manage and organise tasks across different services. 

The Gemini 3.5 Flash will be rolling out today, across all of Google's products and APIs, whereas Gemini 3.5 Pro is set to debut in June. Whereas, Gemini Spark will initially roll out to AI Ultra users in the US.

Must read: Google I/O 2026: Google unveils Gemini Omni AI video editing model
 

Google Gemini 3.5 Flash: Performance and capabilities

According to a Google blog post, the new Gemini 3.5 Flash model is said to be "the strongest agentic and coding model," and excels in coding, agentic capabilities, and multimodal understanding. The model is designed to perform complex AI tasks much faster and cheaper than other advanced AI models. It is said to be perfect for developers and businesses to use as it can quickly plan and organise tasks, build and improve solutions iteratively, and handle practical business or software-related work.

Gemini 3.5 Flash can help users develop new apps, maintain or update existing codebases, and help prepare financial documents. In terms of benchmarks, the new AI model surpassed Gemini 3.1 Pro in coding, reasoning, and multimodal understanding tests, including Terminal-Bench 2.1 with a score of 76.2%, GDPval-AA at 1656 Elo, MCP Atlas at 83.6%, and CharXiv Reasoning at 84.2%. 

The 3.5 Flash is now the default model for the Gemini app and AI Mode in Search globally. However, the Gemini 3.5 Pro is said to debut next month. During the keynote address, Google highlighted that the Pro model will be a powerful version of Flash, bringing powerful capabilities across its products and platforms.

Google also showcased Antigravity 2.0, describing it as an advanced agentic coding system capable of creating the core components of an operating system in half a day. The company claims that it launched 93 separate “sub-agents,” generated 2.6 billion tokens, and completed the core framework of an operating system in roughly 12 hours.

Google Gemini Spark

Gemini Spark was launched as a new agentic personal assistant that runs on Gemini 3.5 and uses the Antigravity harness. The AI agent connects Workspace tools like Gmail, Docs, Slides and more, working as a “cloud-based agent.” It does not need the user’s device to stay active all the time, as it processes tasks in the background even after the user closes their laptop or locks their phone.

With Gemini Spark, users can create complete workflows, such as synthesising meeting notes across emails and chats, creating polished Google Docs with their findings, and even drafting an email on their behalf. Furthermore, users can also set recurring tasks or triggers, and teach the agent a new skill.

The Gemini Spark will be rolling out to limited testers this week and as a beta version to U.S. Google AI Ultra subscribers next week.