Google I/O 2026: Gemini 3.5 Flash Redefines AI

Google I/O 2026: A Revolution Named Gemini 3.5 Flash

At the annual Google I/O conference held on May 19, 2026, Google confirmed the rumors by unveiling the Gemini 3.5 series. At the heart of this announcement, the Flash 3.5 model was presented as a major advancement for AI agents and coding, surpassing its predecessors in terms of speed and cost.

Unprecedented Performance

Google's CEO, Sundar Pichai, emphasized that Gemini 3.5 Flash is not only faster but also more economical than the previous model, Gemini 3.1 Pro. This new model is now globally accessible through the Gemini app and in the AI mode of Google Search, becoming the default model for users.

Developers can also leverage this model via the Gemini API in Google AI Studio and Android Studio. Designed to be lightweight and fast, Gemini 3.5 Flash can produce up to four times more tokens per second than several of its competitors, while being two to three times cheaper in certain contexts.

Superiority in Testing

The performance of Gemini 3.5 Flash is impressive: it scored 76.2% on Terminal-Bench 2.1 in software development, surpassing the 70.3% of Gemini 3.1 Pro. In agent tasks, it achieved 1656 points on GDPval-AA Elo, compared to 1314 for its predecessor. Additionally, it recorded 83.6% on MCP Atlas for tool usage. Google claims that this model nearly rivals Claude Opus 4.7 Max on the Artificial Analysis Intelligence Index, while generating responses about twelve times faster.

Enhanced Security

Despite its increased capabilities, Google assures that Gemini 3.5 adheres to strict security standards thanks to its Frontier Safety Framework. This framework aims to protect against cyberattacks and harmful content. Interpretative analysis tools are used to examine the AI's internal reasoning before responses are transmitted to users.

Introduction of Gemini Spark

Google also introduced Gemini Spark, a personal AI agent based on Gemini 3.5 Flash. Operating on the Google Cloud infrastructure, Spark natively integrates with Google Workspace, facilitating interaction with tools like Gmail and Google Docs. Users can interact with Spark via a dedicated Gmail address, and the agent uses Chrome to manage web tasks.

Josh Woodward, vice president of Google Labs, illustrated concrete use cases for Spark, such as retrieving information to automatically draft reports or email drafts. Some small businesses are already using Spark to monitor their inboxes and avoid missing customer requests.

Finally, Gemini Spark supports connections with third-party services via MCP, and Google plans to expand integrations. Currently, the rollout is limited to a small group of testers, with a broader beta expected next week for American subscribers to Google AI Ultra, a premium service priced at $100 per month.

Android Halo Feature

Android users will also benefit from a new feature called Android Halo, designed to track the actions of the AI agent in real-time on mobile devices.