Gemini 4 is almost ready, says new Google DeepMind chief

In his first media appearance leading Google’s DeepMind division, Koray Kavukcuoglu told The Information that Gemini 4 is in its refinement stage and that Google aims to launch “much earlier” than the end of 2026.

Google hasn’t shipped a new flagship since the Gemini 3 series in November 2025. In that gap, OpenAI’s GPT-6 and Anthropic’s Mythos models launched and, per The Verge’s framing, outperform Gemini 3 — leaving Google behind on the frontier scoreboard. A promised Gemini 3.5 Pro update Pichai floated for June after I/O never rolled out; Kavukcuoglu said the company “took a little bit of a step back” to push faster Flash models instead.

Leadership context: Demis Hassabis stepped down in August after saying “great progress” was underway on Gemini 4. The new boss is now putting a clock on the release.

We’re treating the quote as a pacing signal, not a ship date.

“As soon as possible” plus “early post-training output” because “we see the results and we are excited” is lab-speak for: the base model is good enough to start the public iteration loop. That’s healthier than another year of Flash-only appetizers while rivals redefine the default enterprise model. Kavukcuoglu’s certainty line — “it’s a certainty that we are always gonna be at the frontier” — is confidence theater until a build ships with independent evals.

Useful AI needs Google in the race. Search distribution, Android, Cloud, and DeepMind research are how capable models become everyday tools instead of API boutique products. A long flagship drought doesn’t just bruise prestige; it starves product surfaces that only move when the frontier model moves.

The downside of “fast-paced iterations” after a leadership shakeup is the same calendar war we keep criticizing elsewhere: announce early, refine in public, let safety and quality reviews compress to match the competitive clock. Skipping 3.5 Pro to chase Flash speed may have been right for latency products and wrong for buyers who needed a serious mid-cycle upgrade.

I’m watching for an actual early post-training release window measured in weeks, not vibes — and whether Gemini 4’s first public scores close the GPT-6 / Mythos gap or merely narrow it. “Much earlier than end of year” still leaves a wide runway. Ship the model; then talk frontier certainty.

The Verge’s Jess Weatherbed, Sep 24, 2026, reporting Koray Kavukcuoglu’s Information interview on Gemini 4’s refinement-stage timeline, the post–Hassabis leadership change, and Google’s missed 3.5 Pro beat after prioritizing Flash models.