Google DeepMind's New Chief Says Gemini 4 Could Ship Much Sooner Than Planned

Koray Kavukcuoglu, the new head of Google DeepMind, used his first public remarks in the job to say Gemini 4 has entered early post-training and could launch well before the end of 2026. Alphabet's stock wobbled on the news, capping a rough week against Meta's fresh AI hardware push.

Koray Kavukcuoglu had never spoken publicly as Google DeepMind's chief until this week. He picked a good week to start. Speaking at The Information's AI Agenda Live Summit, Kavukcuoglu said Gemini 4 has moved into early post-training and that Google wants it out much sooner than the late-2026 window most people had assumed.

"Our intention is to roll out an early post-training version as soon as possible, because we've already seen promising results and are very excited," Kavukcuoglu said, according to The Information. He added that Google plans to "continue the fast-paced iterations" once the model is out. That's a real shift in tone. Pretraining and post-training are different stages of building a model, and reaching post-training this early suggests the base model is largely locked and the remaining work is tuning it for release.

Sundar Pichai first confirmed the project on July 21, when Google's Gemini team wrote, in the post announcing Gemini 3.6 Flash, that it had "already started our most ambitious pre-training run yet, for Gemini 4." Two days later, on Alphabet's second-quarter earnings call, Pichai named the model directly and called it "very ambitious," pointing to coding and agentic coding as the areas Google most needs to fix. He also said Google wants to move toward an almost monthly release cadence for Gemini, a sharp break from its historically slower pace.

You can see why the urgency matters. Gemini has fallen behind in the one category that now drives most enterprise AI spending: coding. GPT-5.5 has become the default choice for coding workloads in 2026, with Anthropic's Claude Code the main alternative, and Google is largely absent from that conversation, according to reporting from Axios. One Google employee told Axios the company is simply "behind" on model capability compared with its rivals, even as Google keeps an edge in owned infrastructure and compute.

GOOGL fell 3.8% in Wednesday's session, part of a nearly 5% slide over two trading days driven largely by investor worry over Meta's Muse gaining traction with consumers. Meta unveiled Muse Charm, a two-inch standalone AI hardware device built around its Muse assistant, at Meta Connect on September 23, alongside new audio-only smart glasses due in October and a slimmer VR headset planned for next spring. After Kavukcuoglu's remarks landed, GOOGL shares recovered slightly, trading up about 0.6% in after-hours activity, according to Yahoo Finance.

That's a small bounce off a real dip. But it shows how closely investors are now tracking Gemini's release calendar as a proxy for whether Google can hold its ground against OpenAI, Anthropic and Meta all at once.

For developers and enterprise buyers, the practical question is simpler: does an earlier Gemini 4 change what you build on right now? Probably not yet. Google still hasn't named a release date, and "much earlier" from a DeepMind chief is not the same as a shipping model. Post-training can take weeks or months depending on how much safety testing and fine-tuning Google wants to do before a public launch. OpenAI and Anthropic have both leaned on rapid iteration cycles this year, shipping incremental model updates every few weeks rather than waiting for a single flagship moment. If Google wants to compete on more than benchmark scores, matching that cadence, not just the eventual scale of Gemini 4, is the harder problem.

Kavukcuoglu didn't give a date. He gave a signal. And for a company that just watched two straight trading sessions of selling on competitive fear, a signal was apparently worth sending in public, on stage, in his very first appearance as DeepMind's chief.

Also read: OpenAI Fires Contractors Caught Using ChatGPT to Rate ChatGPT's Answers • Microsoft Slashes Copilot Prices for Its Biggest Buyers as a Super App Looms • Qualcomm's New Snapdragon Chip Runs a 30 Billion Parameter AI With No Cloud

This article is posted in AI News, check it out for more related stories.