Google's Gemini Flash winning customers

Business Insider reported on May 15 that Google’s Gemini 3 Flash model has moved ahead of Anthropic on Vercel’s AI Gateway, a data point that puts a concrete numberless but visible usage shift behind Google’s AI push before its annual developer conference next week. Vercel CEO Guillermo Rauch highlighted the change using gateway token traffic, which tracks how much model output and input developers are sending through the service. (businessinsider.com) The crossover matters because Vercel’s gateway is used by startups, software companies and enterprise product teams building chatbots, coding tools, search products and copilots, according to the report. Google I/O 2026 begins May 19 in Mountain View, California, where the company has already said Gemini updates will be part of the event. (blog.google) ### What exactly changed on Vercel’s AI Gateway? Vercel’s AI Gateway showed Anthropic models leading in March, before Google’s Gemini 3 Flash moved into first place in early April and remained there through mid-May, according to Business Insider’s description of a chart shared by Rauch. The measure cited was token traffic rather than revenue. Guillermo Rauch, Vercel’s chief executive, singled out Gemini Flash rather than Google’s largest models. Vercel markets AI Gateway as a single interface for routing requests to “hundreds of AI models,” which makes its traffic a useful read on what developers are choosing for live products, though it is still one platform’s view rather than a marketwide tally. (businessinsider.com) ### Why would a smaller Google model win more traffic? Gemini Flash is positioned as a faster, lower-cost class of model, and the Business Insider report said that profile is helping it win common production workloads. The customers described in the report are teams running user-facing AI features where latency and cost matter on every request. (businessinsider.com) Vercel’s gateway setup also favors mix-and-match usage. The company says developers can route text, image and video requests through a centralized interface, which means a cheaper model can handle routine tasks while a more capable model is reserved for harder prompts or fallback cases. That routing pattern is an inference from how gateway products are typically used, but Vercel’s own product description confirms the multi-model architecture that makes it possible. (businessinsider.com) ### Does this mean Anthropic is losing ground everywhere? Business Insider’s report did not say that. The clearest claim in the sourced material is narrower: Google passed Anthropic in token traffic on Vercel’s AI Gateway, while Anthropic had led there in March. (businessinsider.com) One follow-on account of Rauch’s remarks said Anthropic still led on revenue share on the platform even after Google moved ahead on raw token usage, but that detail did not appear in the primary Business Insider excerpt available here. Because Vercel has not published a full public market breakdown in the material reviewed, the gateway data should be read as a platform-specific indicator, not a complete ranking of enterprise AI demand. (vercel.com) ### What is Gemini Spark, and how solid is that report? 9to5Google reported on May 14 that Google is working on an AI agent called “Gemini Spark” inside the Gemini app. (businessinsider.com) The publication said the finding came from an “APK Insight,” based on decompiling app code uploaded to the Play Store. 9to5Google said the code pointed to a more advanced agent capability that could handle tasks such as inbox cleanup, meeting briefs and custom news digests. The outlet also cautioned that code strings do not guarantee a feature will ship and said its interpretation could be imperfect. That makes Spark a pre-announcement signal rather than a confirmed launch. (biztechweekly.com) ### What has Google confirmed for next week? Google said in a Feb. 17 post that I/O 2026 will run May 19-20 at Shoreline Amphitheatre in Mountain View and online, and that the event will cover “latest AI breakthroughs” and updates “from Gemini to Android and more.” The company has not, in the official material reviewed, named Gemini Spark. (9to5google.com) May 19 is the next hard date in the story. Google’s keynote starts the first day of I/O 2026, and any formal announcement on Gemini models, agents or developer tooling would come from that stage or the company’s event materials. (blog.google) (9to5google.com)