Google pushes Gemini 3.5 Flash

Google used its May 19 I/O 2026 conference to push Gemini further into enterprise software with Gemini 3.5 Flash, a new model the company says is built for coding and agentic work rather than chatbot-style interaction. Google said the model combines “frontier intelligence with incredible speed” and is designed for workflows where systems need to plan, execute and iterate across multiple steps. The release puts speed at the center of Google’s pitch. In its announcement, Google said Gemini 3.5 Flash outperforms Gemini 3.1 Pro on almost all benchmarks while running four times faster than other frontier models. Koray Kavukcuoglu, DeepMind’s chief technologist, told reporters the model is aimed at coding, agentic tasks and multimodal reasoning, according to TechCrunch. (blog.google) That framing matters because Google is not selling 3.5 Flash mainly as a bigger general-purpose model. Google’s own materials describe it as the first model in a new family that combines “frontier intelligence with action,” while I/O product pages tied the launch to Antigravity, Google’s agent-first development platform. (blog.google) ### Why did Google emphasize speed instead of scale? Google said agentic workflows need a high-speed model because multiple agents may run in parallel on longer tasks. Kavukcuoglu said 3.5 Flash is four times faster than other frontier models and that Google also built an optimized version that is 12 times faster at the same quality level, according to TechCrunch. (blog.google) Google’s benchmark claims were also tailored to that pitch. In its model post, the company said Gemini 3.5 Flash beat Gemini 3.1 Pro on coding and agentic tests including Terminal-Bench 2.1, GDPval-AA and MCP Atlas, while also posting strong multimodal reasoning scores. ### What is Google saying this model is supposed to do? (techcrunch.com) Google described Gemini 3.5 Flash as its “strongest agentic and coding model yet.” The company said the model is intended for real-world action systems, not only question answering, and linked it to tools that let developers build agents that can complete tasks inside software environments. (blog.google) TechCrunch reported that Google presented the model as capable of independently executing coding pipelines and managing research projects, and said internal tests included building an operating system from scratch. Those examples were part of Google’s argument that AI systems are moving from assistants toward agents that can handle longer chains of work. (blog.google) ### Where does enterprise adoption get tested? InfoWorld reported that Google is pushing Gemini 3.5 Flash deeper into enterprise workflows, especially where companies want to embed agents into software used for support, automation and other operational tasks. The publication said Google’s message is that the model is faster and better suited to coding and agentic work. (techcrunch.com) InfoWorld also said analysts cautioned that enterprise value will depend less on launch-day benchmark claims than on reliability in live workflows. That means buyers are likely to judge the model by how consistently it performs inside business processes rather than by raw model-size comparisons. (infoworld.com) ### How does this fit into Google’s broader I/O push? Google tied Gemini 3.5 Flash to a broader I/O 2026 product cycle built around agents. In its conference roundup, the company said Gemini 3.5 Flash was the first model in its latest family and paired it with Antigravity and other developer tools meant to move users “from prompts to action.” (infoworld.com) Google also said the Gemini app now reaches more than 900 million monthly users across 230 countries and more than 70 languages, giving the company a large distribution base as it rolls out newer models and agent features. ### What comes next for buyers and developers? Google said developers can build with Gemini 3.5 Flash through the Gemini API, AI Studio and the company’s broader developer stack announced at I/O 2026. (blog.google) The next test will come as enterprises put the model into production workflows and measure latency, reliability and task completion in live systems. (blog.google 1) (blog.google 2)