Google makes Gemini 3.5 Flash the default across Search and Gemini app, unveils Spark agent

Google on Tuesday introduced the Gemini 3.5 model family and immediately made Gemini 3.5 Flash the default AI model in both the Gemini app and AI Mode in Google Search worldwide, giving the company’s latest system an instant global footprint across consumer products.

The rollout, announced in a Google blog post published May 19 as part of Google I/O 2026, signals that Google is pushing beyond chatbot-style AI responses and toward models designed to carry out multistep tasks. “Today, we’re introducing Gemini 3.5, our latest family of models combining frontier intelligence with action,” the company said. Google added that “3.5 Flash is available today to billions of people globally.”

Google said Gemini 3.5 Flash is available starting Tuesday in the Gemini app and AI Mode in Search, as well as for developers through Google Antigravity and the Gemini API in Google AI Studio and Android Studio. Businesses can access it through Gemini Enterprise Agent Platform and Gemini Enterprise, Google said. The company said it is also working on Gemini 3.5 Pro, a higher-end version already being used internally that is expected to roll out next month.

The announcement also included Gemini Spark, which Google described as a personal AI agent powered by 3.5 Flash that runs continuously in the background. Google said Spark is rolling out to trusted testers Tuesday and is scheduled to enter beta next week for Google AI Ultra subscribers in the U.S. The product gives a clearer picture of how Google intends to use the new model in more persistent, action-taking assistants rather than one-off chat sessions.

Google framed Gemini 3.5 Flash as a faster, more capable system for coding and other agentic work, though the performance figures released Tuesday are Google’s own and have not yet been independently validated in the company’s announcement materials. Google said 3.5 Flash outperforms Gemini 3.1 Pro on coding and agentic benchmarks, including a score of 76.2% on Terminal-Bench 2.1, 83.6% on MCP Atlas and 84.2% on CharXiv Reasoning. The company also said the model reached 1656 Elo on GDPval-AA and is “4 times faster than other frontier models” by output tokens per second. Google illustrated the model’s intended uses with examples such as synthesizing a paper and coding a playable game in six hours, refactoring legacy code to Next.js through the Antigravity harness, and generating interactive user interfaces and animations.

On safety, Google said Gemini 3.5 was built under its Frontier Safety Framework, the company’s public system for evaluating risks from its most advanced models. “Gemini 3.5 was developed in accordance with our Frontier Safety Framework,” Google said. The company said it strengthened safeguards related to cyber and CBRN risks — shorthand for chemical, biological, radiological and nuclear threats — and added safety training, mitigations and interpretability tools meant to help inspect how the model reasons through tasks.

Google also highlighted early enterprise use and pilots for 3.5 Flash, naming companies including Shopify, Salesforce and Databricks. That matters because Gemini 3.5 is not being introduced as a standalone research model. Google is deploying it at the same time across consumer search, its flagship chatbot, software development tools and enterprise automation products.

Gemini is Google DeepMind’s flagship multimodal model family, built to handle different kinds of inputs such as text, images and code. Within that lineup, Flash is the lower-latency, efficiency-focused tier, while Pro is positioned for deeper reasoning. By making 3.5 Flash the default model immediately and tying it to products aimed at consumers, developers and businesses, Google is signaling that its next phase for AI is not just better answers, but systems meant to do work.