Gemini 4 Rumors Are Mounting as Google Races to Catch Up
Trending Topics Jakob Steinschaden ● Covered by 2 sources
Google says Gemini 4 is coming sooner than expected, with leaks already circling. That matters because its last real flagship is old news and rivals are pulling ahead.
Based on reporting by Trending Topics, Jakob Steinschaden — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google is trying to get a new flagship model out the door, and Gemini 4 may arrive faster than many expected. Koray Kavukcuoglu, who leads Google DeepMind, said at The Information’s AI Agenda Live Summit that the model is already in post-training and should launch “much earlier” than the end of the year. Google has not said much else officially, but the direction is clear: this one is being tuned for safety and reliability, and the company wants an early version out quickly before tightening it up.
The pressure here is partly of Google’s own making. Gemini 3.5 Pro, which was announced for I/O in May, never shipped after reports that its coding performance was not good enough. Google leaned on cheaper Flash models instead, including Gemini 3.7 Flash in August and Gemini 3.8 Flash in early September. Its last true flagship, Gemini 3 Pro, goes back to November 2025.
Leaks are filling in the blanks, though they should be treated carefully because they do not all agree. TestingCatalog says Google has been testing Gemini 4 Pro checkpoints under the code name “Argon” since mid-September. Those reports claim much longer outputs, with up to 256,000 tokens in one response, and say the model can take anywhere from 2.4 to 20 minutes in its highest reasoning mode. Developers on X are also talking up the model’s websites and SVG graphics, saying earlier design flaws have finally been fixed.
Then there’s the messier stuff. Estimates for the context window range from 1.5 million tokens to 2 million to more than 10 million, but none of those figures is backed up. An unnamed model called “gemini-3.8-flash” also showed up on LMArena in mid-September, which some people think was a disguised Gemini 4 variant. And TestingCatalog says the Gemini desktop app is being rebuilt with avatars, plugins, and widgets for Gmail and Calendar. October is being thrown around on X, while TestingCatalog points to November or December based on traces in the app code.
All of this would matter less if Google were comfortably ahead. It isn’t. Artificial Analysis puts its best model, Gemini 3.8 Flash, at 41 points, while Anthropic’s Claude Opus 5.5 leads with 58. Google is running a cheap Flash model against rivals’ flagship systems, and that is not a great look for a company with Gemini, Android, Chrome, Search, and its own A.I. chips. The race is tight, and Google knows it.
My take — AI-written commentary, not fact-checked reporting
Google keeps acting like the future will save the present, which is a classic big-tech hobby. Shipping a flashy new model after letting the flagship gap open is not strategy, it’s cleanup with better branding. The industry has a real habit of calling every delayed model “the frontier” right up until the benchmark chart gets awkward.
Read more about this at: Trending Topics
