Skip to main content

Google's Gemini 4 Is Almost Here—and It's Coming Sooner Than You Think

Google's Gemini 4 Is Almost Here—and It's Coming Sooner Than You Think

At The Information's AI Agenda Live event on September 23, Demis Hassabis, the head of Google DeepMind, dropped a bombshell: Google's next flagship model, Gemini 4, is already in the "early post-training" phase. That's the stage where the base model gets fine-tuned and its behavior becomes more reliable. In plain English, it's getting its final polish before hitting the big stage.

Hassabis didn't mince words—he wants Gemini 4 out "well before the end of the year." And why the rush? Because the early results are looking so good that the team is genuinely excited. "We saw the results and were very excited," he said. After launch, expect rapid iterations. This will be Google's first next-gen flagship since the Gemini 3 series last year. In February, they rolled out Gemini 3.1, then a bunch of smaller, cheaper Flash models. But now, it's all eyes on Gemini 4.

What Happened to Gemini 3.5 Pro?

Remember when Google CEO Sundar Pichai promised a major Gemini 3.5 Pro update at I/O in May? Well, June came and went—and nothing. Hassabis explained that after Gemini 3 and 3.1, Google decided to "take a step back" and focus on Flash models instead. "For us, the most important thing at that time was to accelerate learning as much as possible," he said. He didn't confirm whether 3.5 Pro is officially canceled, but he made it crystal clear: Gemini 4 is the priority now.

Already in the Wild—Internally, At Least

So, is Google worried about falling behind competitors? Hassabis shot that down instantly: "100% confidence in the team." He added, "Our team is incredibly strong, and I have the greatest trust in them; we will always be at the forefront of technology." Right now, Gemini 4 is undergoing safety testing and protective mechanism development. Google engineers are already using it internally to run the Antigravity AI programming tool—so it's not just a lab experiment.

The AGI Question? Hassabis Isn't Biting

Unlike some rivals who love to debate whether we've achieved AGI, Hassabis sidestepped the topic. "Discussing whether AGI has been achieved is not the right direction," he said. "The real issue is whether we can build trustworthy agents." That's a refreshingly grounded take in a field often obsessed with buzzwords.

Key Points

  • Gemini 4 is in early post-training and could launch well before the end of the year.
  • Google skipped Gemini 3.5 Pro to focus on accelerating learning with Flash models.
  • Internal use is already happening, with engineers running the Antigravity AI programming tool.
  • Safety testing and protective mechanisms are currently underway.
  • Hassabis has 100% confidence in his team and says Google will stay at the forefront.
  • No AGI hype here—he'd rather focus on building trustworthy agents.