Google is bringing forward the release of its next-generation AI model, Gemini 4, to much earlier than the end of this year, a tech publication reported on Sept. 25, citing The Information.

The report said Koray Kavukcuoglu (코라이 카부쿠오글루), a senior executive at Google DeepMind, disclosed the plan at The Information's "AI Agenda Live Summit."

Kavukcuoglu said Gemini 4 would come out "much sooner" than year-end. Some Google fans are predicting an October launch, but that is still speculation, the tech publication said.

Google DeepMind said Gemini 4 is in a post-training stage. That means the training work to build the base model using large amounts of data has been completed. Post-training often involves fine-tuning and reinforcement learning, and frequently uses human feedback. It is not known whether the Gemini 4 Pro model will come first or the Flash model will come first.

Google released Gemini 3 Pro late last year and has since appeared to fall behind rivals in large language models. High-performance LLMs were led by OpenAI and Anthropic, while Google focused on smaller, more efficient models, the tech publication said.

Until last year, Google was considered the most promising AI company.

It was the only company that could achieve full vertical integration by having infrastructure, data, applications and models. Even hyperscalers such as Microsoft and Amazon did not have those conditions.

Still, Google did not produce impressive results. It released Antigravity and Gemini CLI this year, but they did not establish themselves as coding tools that could take on OpenAI Codex or Claude Code.

Against that backdrop, Meta and SpaceX AI are also showing they could push Google out of the No. 3 AI spot in the United States, the tech publication said. SpaceX AI's Grok and Meta's Muse Spark perform better overall than the top-end Gemini Flash model.