Google Gemini 4 Is Almost Ready: DeepMind Chief Shares the Latest Status
Originally published in Turkish. Read the original →
The head of Google DeepMind announced that pre-training of the flagship AI model Gemini 4 is complete and that it will be released before the end of the year.
Google has started the countdown for Gemini 4, its new flagship model in the generative AI arena. Koray Kavukçuoğlu, the Turkish researcher and executive who took over the leadership of Google DeepMind in August 2026, shared critical details about the model's development process publicly for the first time. Speaking at the AI Agenda Live Summit organized by The Information, Kavukçuoğlu announced in his first media statement since taking the DeepMind seat that the core pre-training stage of Gemini 4 has been completed and that the system is now in its final preparation phase.
Google aims to make this flagship model, which the tech world is eagerly awaiting, generally available well before the end of 2026.
Pre-Training Complete, Moving to the Post-Training Stage
According to the information Kavukçuoğlu shared, Gemini 4 has completed the intensive pre-training process carried out on massive datasets. The AI model is currently at the beginning of the post-training stage. In this critical phase, the model is being carefully optimized through reinforcement learning from human feedback (RLHF), logical reasoning, factual accuracy, safety protocols and systemic consistency filters.
Emphasizing that the DeepMind teams are very pleased with the current test and trial results, Kavukçuoğlu said that the fast iteration cycle is continuing steadily to cement the model's reliability and high performance. Although the company has been cautious about the timeline and has not given an exact date, it was noted that the work is progressing on the intended course.
The Gemini 3.5 Pro Delay and the Flash Series Strategy
As you may recall, the tech giant launched the Gemini 3 series around November 2025. Then, at the Google I/O conference in May, Google CEO Sundar Pichai announced that a comprehensive interim update for Gemini 3.5 Pro would be released in June. However, that big Pro version never reached users.
Kavukçuoğlu also shed light on the strategic decision behind this delay. The veteran executive said the company had deliberately taken a step back at the time and shifted resources to Flash models, which have lower latency, are extremely efficient in terms of compute and offer cost advantages for developers. In line with this strategy, the focus turned to optimized models, while the main flagship push was saved directly for the Gemini 4 architecture.
A New Balance of Power Against OpenAI and Anthropic
Competition in the AI market is moving at a harsher pace than ever. Strong model updates from OpenAI and Anthropic in recent months had increased the flagship performance pressure on Google. Although the Flash models won developers' approval in terms of speed and operational cost, there was still an expectation of a top-tier model for complex reasoning, autonomous software development and multi-step agent workflows.
Gemini 4 is set to combine Google's deep experience in multimodal architecture, deep reasoning capabilities and autonomous system integrations under one roof. Shaped by Google's TPU chip infrastructure and DeepMind's research power, the new model could reshuffle the cards in the AI leadership race before the year is out.
