On May 19, Google unveiled the Gemini 3.5 series at Google I/O 2026; the initial model, 3.5 Flash, is now available worldwide. According to Google, it outperforms Gemini 3.1 Pro in agentic tasks and code generation. It has set new records across key benchmarks such as Terminal-Bench 2.1 (76.2%) and GDPval-AA (1,656 Elo), as well as in the multimodal reasoning test CharXiv Reasoning (84.2%). Its token output speed is four times faster than other leading models, while operational costs for long-term agentic tasks amount to less than half those of competitors. On the Artificial Analysis index, it occupies the upper-right quadrant labeled “cutting-edge intelligence × ultra-fast responsiveness.” The 3.5 Flash model is already integrated into the Gemini app, Google Search’s AI Mode, Google AI Studio, Android Studio, and the enterprise-focused Gemini Enterprise Agent Platform. Meanwhile, the larger flagship model, Gemini 3.5 Pro, is currently being used internally at Google and is slated for public release in June.
Alongside the launch of 3.5 Flash, Google introduced two new infrastructure tools. First is Google Antigravity, an agent-first development platform enabling parallel deployment of multiple sub-agents to handle large-scale, long-duration workflows. Google demonstrated how two collaborating agents managed to read an AlphaZero research paper and develop a playable game within just six hours. Second is Gemini Spark, a personal AI agent built on top of 3.5 Flash; it operates around the clock to perform digital tasks on users’ behalf. Access to Spark is currently limited to trusted testers, with a beta version scheduled to reach U.S.-based subscribers of Google AI Ultra next week. In practical applications, Shopify leverages parallel sub-agents to forecast growth trends among global merchants, Macquarie Bank utilizes the model to analyze documents spanning over 100 pages to expedite customer onboarding, Salesforce integrates it into its Agentforce ecosystem, and Ramp employs it for multimodal OCR-based invoice recognition. Google notes that the entire Gemini 3.5 series was developed under its Frontier Safety Framework, which includes dedicated explainability tools designed to scrutinize internal model reasoning prior to generating responses.