Google readies Gemini rival before I/O

Google has not publicly confirmed a new Gemini flagship model ahead of its annual I/O conference, but a cluster of leaks, app findings and outside reports has sharpened attention on what the company may show next week. Google said in February that I/O 2026 will run on May 19 and May 20 at Shoreline Amphitheatre in Mountain View, California, and that the event will cover “latest AI breakthroughs” spanning Gemini, Android and other products. May 12 and May 13 reports from outside publications described hidden Gemini features and unreleased model names inside Google software, including references tied to Gemini Live and a possible “Omni” video system. Those reports stop short of proving a full new flagship model, but they add to signs that Google is preparing a broader Gemini update cycle around I/O. (blog.google) OpenAI’s release of GPT-5.5 on April 23 has raised the competitive stakes for Google before the conference. OpenAI described GPT-5.5 as its “smartest and most intuitive” model and expanded API availability on April 24, putting a fresh benchmark in front of rivals weeks before Google’s developer event. ### What has Google actually said before I/O? Google’s official position is narrower than the online speculation. (forbes.com) The company’s I/O announcement says developers should expect updates “from Gemini to Android and more,” but it does not name a new Gemini generation or give product-level details. May 12’s Android Show offered one concrete sign of Google’s current direction. (openai.com) Google introduced “Gemini Intelligence,” a set of proactive Android features that automate multi-step tasks, and separately said Chrome on Android will add Gemini-powered tools in late June for eligible U.S. users. Those announcements show Google already moving Gemini deeper into consumer products before the main keynote. (blog.google) Google DeepMind’s current public model lineup also matters here. The company’s Gemini pages now promote Gemini 3 and Gemini 3.1 Pro as its most capable reasoning-oriented systems, with DeepMind saying Gemini 3 Pro improves solved benchmark tasks by more than 50% over Gemini 2.5 Pro. ### Where did the new-model talk come from? Forbes reported this week that hidden Gemini Live models were visible ahead of I/O, suggesting Google has server-side testing infrastructure in place for multiple voice AI variants. (blog.google) Gadgets 360 and other outlets separately pointed to a leaked “Gemini Omni” video model in testing before the conference. (deepmind.google) Those reports describe product clues, not a formal Google launch plan. Some outside sites have gone further, speculating about names such as Gemini 3.2 or 3.5, but those claims are not backed by public Google documentation reviewed here. ### Is this really about matching OpenAI on reasoning? OpenAI’s April 23 launch gives the clearest public marker for the rivalry. (forbes.com) The company said GPT-5.5 is a “new class of intelligence for real work,” and its API pricing page shows higher token pricing than GPT-5.4, underscoring that OpenAI sees it as a premium frontier model. Google DeepMind has framed its own recent releases in similar terms, emphasizing reasoning, reliability and agentic behavior. (aixploria.com) DeepMind says Gemini 3 Pro advances “depth, reasoning, and reliability,” while Gemini 3.1 Pro is described as its “most intelligent model yet.” That makes the pre-I/O chatter plausible, even if unconfirmed: if Google shows a new Gemini tier next week, it would arrive less than a month after GPT-5.5. (openai.com) The timing is an inference from the release calendar, not a statement Google has made. ### Why did researcher pay start trending alongside the leak chatter? (deepmind.google) Epoch AI published an analysis on May 13 saying top AI researchers can earn more than $30 million annually while postdoctoral researchers make about $50,000, a gap it described as more than 100-fold. That piece circulated as discussion about Google, OpenAI and other labs intensified ahead of I/O. (blog.google) The pay debate matters because frontier-model releases and talent competition are closely linked, though the strongest compensation claims often rely on private negotiations rather than public filings. Publicly available salary trackers show high compensation for Google research roles, but they do not establish what Google is paying elite model builders on special packages. (epoch.ai) ### What should readers watch on May 19? May 19 is the next hard date in the story. Google says I/O 2026 opens that day at Shoreline Amphitheatre in Mountain View, and the company has already told developers to expect updates across Gemini and Android. Google’s own pre-I/O announcements suggest the keynote will likely connect model improvements to products, not just benchmarks. (6figr.com) The company has already previewed Gemini Intelligence on Android, Chrome integrations due in late June, and new Gemini-centered hardware concepts such as Googlebook. May 19 and May 20 are therefore the next public test of the leak cycle. (blog.google) If Google plans to name a new Gemini model or expand multimodal tools such as video generation, I/O is the venue the company has already designated for its latest AI announcements. (blog.google)