Gemini 4, the upcoming flagship model from Google DeepMind, is in the final adjustment phase and will be launched "much sooner" than the end of the year, according to Koray Kavukcuoglu, who took over the daily operations of the division in August.
In his first public appearance as the new head of DeepMind before The Information, the executive stated that Googleaims to release a post-training version as soon as possible and maintain a rapid pace of updates.
Gemini 4 is undergoing the final adjustment phase
When will Gemini 4 be launched and what did the new head of DeepMind say
The arrival of Gemini 4 would occur after ten months without a flagship model from Google. The last release in that category was the Gemini 3 family, presented in November 2025.
During that period, OpenAI launched GPT-6 Astra and Anthropic introduced its Mythos series. According to specialized media, both surpassed Gemini 3 in reasoning and programming tests.
The launch date of Gemini 4 will also be significant for the artificial intelligence market. The speed of publication may influence prices, capabilities, and the availability of APIs, in a landscape dominated by Google, OpenAI, and Anthropic.
When will Gemini 4 be released and what did the new head of DeepMind say
What changes occurred in Google DeepMind before the launch of Gemini 4
The promise to accelerate releases coincides with an internal reorganization announced by Google in August. Demis Hassabis, co-founder of DeepMind and Nobel Prize winner for AlphaFold, left daily operations to become president of the division and chief scientist of Alphabet, as reported by TIME.
In his place, Koray Kavukcuoglu, who was serving as chief technology officer and had been leading meetings on Gemini for over a year, now holds the position of senior vice president of DeepMind and reports directly to Sundar Pichai.
What changes occurred in Google DeepMind before the launch of Gemini 4
The reorganization was also accompanied by several notable departures:
Jeff Dean, chief scientist of Google, left the company after 27 years along with several colleagues to found Discovery Loop. Google will be a founding investor and cloud infrastructure partner.
John Jumper, co-discoverer of AlphaFold, joined Anthropic.
More than 580 employees signed an open letter asking Pichai to reconsider a contract with the Pentagon, as reported by Insider Monkey.
In a few weeks, Google lost several high-profile figures and concentrated the leadership of Gemini in Kavukcuoglu. The executive stated that the priority is to accelerate publication cycles.
Why Google did not launch Gemini 3.5 Pro and what happened to the 3.x family
Kavukcuoglu acknowledged that Google "took a step back" to focus on the Flash models, designed to offer greater speed and lower costs, instead of moving forward with Gemini 3.5 Pro.
Why Google didn't launch Gemini 3.5 Pro and what happened to the 3.x family
The model had been announced by Sundar Pichai during Google I/O 2026, with a launch planned for June, but it never reached the public. Bloomberg and Axios reported that it did not meet internal goals, while Ars Technica noted that it remains in testing phase.
Meanwhile, Google expanded the 3.x family with three models aimed at different uses:
Gemini 3.6 Flash: achieved 49% on DeepSWE, a programming test, compared to 37% for Gemini 3.5 Flash. It also reached 83% on OSWorld, a computer usage test, and consumed 17% fewer tokens. Its API costs US$1.50 per million input tokens and US$7.50 per million output tokens.
Gemini 3.5 Flash Lite: reaches a speed of 350 tokens per second and is designed for large-scale agents. Its API costs US$0.30 per million input tokens and US$2.50 per million output tokens.
Gemini 3.5 Flash Cyber: is the first version of Gemini focused on cybersecurity.
The model had been announced by Sundar Pichai during Google I/O 2026
How Gemini 4, OpenAI, and Anthropic compete in price and performance
Google's strategy focused on reducing the cost per task, rather than competing solely on the intelligence level of its models. However, its rivals also advanced in reasoning, programming, and efficiency.
Anthropic introduced Claude Opus 5.5, with an operating cost 40% lower and a speed 30% higher than Opus 5, according to the company.
OpenAI, for its part, launched GPT-6 Sol, with prices of US$2 per million input tokens and US$10 per million output tokens, and GPT-6 Luna, with rates of US$0.10 and US$0.50, respectively. The company applied a 50% reduction compared to the previous generation.
