Google Targets Gemini 4 Release Well Before 2026 Year-End
Gemini 4 has entered post-training and is being tested on Google’s Antigravity coding platform as the company seeks to close a 10-month flagship-model gap.
Google is targeting a Gemini 4 release well before the end of 2026 after the flagship artificial intelligence model entered post-training, putting pressure on Google to close a 10-month gap between major releases.
Google DeepMind senior vice president Koray Kavukcuoglu said Gemini 4 is already being tested on Antigravity, Google’s coding platform. The company plans to publish early post-training results and then continue improving the model through rapid iterations.
Post-training is the stage after a model’s main development, when it is fine-tuned and adjusted using additional feedback. Google’s latest flagship model, Gemini 3, was released in November 2025.
The Gemini 4 work follows delays involving Gemini 3.5 Pro. Sundar Pichai said in May that the upgraded model would arrive in June, but the June, July and August targets passed without a launch.
Kavukcuoglu said Google moved resources toward faster, smaller Flash models instead of continuing to delay its flagship model. The strategy is intended to get usable post-training results into users’ hands sooner and maintain a faster development cycle.
The company’s leadership structure also changed Aug. 5. Demis Hassabis became chairman of Google DeepMind and chief scientist at Alphabet, while continuing to focus on artificial general intelligence and Isomorphic Labs. Kavukcuoglu took responsibility for model development, frontier AI research and the Gemini app teams, reporting directly to Pichai.
Kavukcuoglu framed the central challenge as building trustworthy intelligent agents rather than determining whether Google had achieved artificial general intelligence. He also expressed confidence that Google remains at the frontier of AI development.
Google’s AI business figures include more than 2.4 million weekly active Antigravity users and about 22 billion tokens processed per minute by its model APIs.
The company’s Gemini 4 post-training work is underway, but the release remains a target, not a completed launch.
