Google is finalizing Gemini 4. New DeepMind head promises release as soon as possible

OpenAI and Anthropic have been releasing one new model after another in recent weeks, while Google last updated its most powerful model in February. That is about to change. According to the new head of Google DeepMind, Gemini 4 is in the final stages of development, and the company wants to release it as soon as possible. Users of the Gemini app can thus look forward to a smarter assistant that could finally catch up with the competition.

What stage is Gemini 4 in?

Koray Kavukcuoglu, who took over leadership of the Google DeepMind division in August, spoke at the AI Agenda Live conference this week. It was his first public appearance in the new role. According to what he told The Information, Gemini 4 is undergoing so-called post-training, which is the phase where the model’s behavior is fine-tuned before release. This includes security tests and setting safeguards. Internally, Google is already testing the model in its Antigravity programming environment.

“Our intention is to release an early version after post-training as soon as possible because we see the results and are happy with them,” Kavukcuoglu stated. No specific date was given, but the release is expected “much earlier” than the end of 2026. According to him, Google wants to continue maintaining the fast pace at which it releases new model versions.

Gemini 3.5 Pro Never Released

Google introduced its last major flagship model in November 2025, when Gemini 3 Pro was released. In February, it was complemented by Gemini 3.1 Pro, and since then, the faster and cheaper Flash series has received updates, most recently Gemini 3.8 Flash in early September. However, Gemini 3.5 Pro, which Google CEO Sundar Pichai announced for June at the May I/O conference, was never released. Kavukcuoglu admitted that the company “backed off a bit” and focused on the Flash models instead.

Google is finalizing Gemini 4. New DeepMind head promises release as soon as possible Adam Kurfürst News Adam Kurfürst News

The new pace comes after leadership changes. Demis Hassabis handed over the day-to-day management of DeepMind to Kavukcuoglu in August and became the division’s chairman and Alphabet’s chief scientist himself. In a letter to employees at the time, he wrote that “great progress” was being made on new models, including Gemini 4.

OpenAI and Anthropic Battled It Out in the Meantime

The competition certainly didn’t sit idle. In early September, OpenAI introduced GPT-6 Astra, its most powerful model to date. Then, on Tuesday, September 22, Anthropic released Claude Opus 5.5, which outperforms even the more expensive Fable 5.1 in several tests and is expected to cost developers about 40% less than Opus 5.

OpenAI responded about an hour later. It added a pair of cheaper models, GPT-6 Sol and GPT-6 Luna, to its Astra, which cost half as much in the API compared to their predecessors from the 5.6 series. Both companies currently have a significant lead over Google.

However, Kavukcuoglu dismisses concerns about falling behind. “I have complete faith in the team,” he said, adding that he believes it is certain that Google will always be at the forefront. He shifts the debate about artificial general intelligence (AGI) elsewhere – according to him, it’s more about whether we can build intelligent agents that can be trusted.

Do you think Gemini 4 can erase the lead of OpenAI and Anthropic?

Sources: The Information, The Verge, 9to5Google