Google unveils Gemini 3.5 Flash at I/O

Google used its May 19 I/O keynote to launch Gemini 3.5 Flash, the first model in its new Gemini 3.5 family, and said it was built for coding and agentic workflows. The company said the model outperforms Gemini 3.1 Pro on benchmarks including Terminal-Bench 2.1, GDPval-AA and MCP Atlas while running four times faster than other frontier models. Google also said Gemini 3.5 Flash is already live in the Gemini app, AI Mode in Google Search, Google AI Studio and its Antigravity development platform. The launch adds a complication to Google’s pitch on cost. Google’s developer pricing page lists Gemini 3.5 Flash at $1.50 per million input tokens and $9.00 per million output tokens, below Gemini 3.1 Pro’s listed $2.00 and $12.00 rates on a per-token basis. But The Decoder, citing analysis from Artificial Analysis, reported that higher token use in agent tasks can push total benchmark costs above Gemini 3.1 Pro and as high as 5.5 times the predecessor model. (blog.google) ### If Flash is supposed to be the fast model, what exactly did Google say improved? Google DeepMind said Gemini 3.5 Flash is its “strongest agentic and coding model yet.” In the company’s launch post, Google said the model scored 76.2% on Terminal-Bench 2.1, 1656 Elo on GDPval-AA, 83.6% on MCP Atlas and 84.2% on CharXiv Reasoning, which it described as multimodal understanding. (ai.google.dev) Google also said the model produces output at four times the speed of other frontier models and is designed for “complex long-horizon tasks.” In the same post, the company said it is using the model with an updated Antigravity harness for multi-step workflows and collaborative subagents. ### Why are outside testers saying the bill can still rise? (blog.google) Artificial Analysis, as described by The Decoder on May 20, found that Gemini 3.5 Flash’s token prices had tripled from Gemini 3 Flash levels to $1.50 per million input tokens and $9.00 per million output tokens. The report said that while those rates remain below Gemini 3.1 Pro on paper, the newer model consumed enough tokens in agent-style benchmarks to make full-task costs nearly twice Gemini 3.1 Pro in some tests and 75% higher on agent tasks. (blog.google) That distinction matters because Google’s own pricing page charges not only for tokens but also for add-ons such as grounding with Google Search and Google Maps after a monthly free allotment. Google lists those search and maps grounding charges at $14 per 1,000 queries after 5,000 free prompts per month shared across Gemini 3 models. (the-decoder.com) ### Where is Google putting the model first? Google said Gemini 3.5 Flash is available “today to billions of people globally” through the Gemini app and AI Mode in Google Search. The company also said developers can access it through the Gemini API in Google AI Studio and Android Studio, while enterprises can use it in Gemini Enterprise Agent Platform and Gemini Enterprise. (ai.google.dev) Josh Woodward, Google’s vice president for Google Labs, Gemini app and AI Studio, said in a separate product post that more than 900 million people now use Gemini monthly across 230 countries and more than 70 languages. In that post, Google described Gemini 3.5 Flash as part of a broader update to make the app more agentic. (blog.google) ### What else did Google pair with the model launch? Google introduced Gemini for Science on May 19 as a collection of research tools and experiments tied to the same I/O push. Pushmeet Kohli and Yossi Matias said the package includes Hypothesis Generation, Computational Discovery and Literature Insights, built with systems including Co-Scientist, AlphaEvolve, Empirical Research Assistance and NotebookLM. (blog.google) Google also said Gemini 3.5 Pro is coming next month. In the Gemini 3.5 launch post, the company said the Pro model is already being used internally and will be rolled out later in June 2026. (blog.google 1) (blog.google 2)