Google Confirms Gemini 3 AI Model Release Timeline: Tipped to Offer Improved Reasoning

Gemini 3 artificial intelligence (AI) model will be released later this year, confirmed Google CEO Sundar Pichai on Wednesday. The confirmation came during his Q3 2025 earnings call, where the main highlight was the company hitting the milestone of $102.3 billion (roughly Rs. 9.06 lakh crore) in revenue in the quarter. The purported upcoming Gemini model is now confirmed to arrive seven months after the company released the Gemini 2.5 Pro model. Notably, the Gemini 2.5 family of AI models were expanded to all users globally in July.

Gemini 3 AI Model Confirmed by Google CEO

Over the last couple of months, the social media platforms have been filled with rumours around Gemini 3. One such rumour, earlier this month, claimed that the purported model could be released on October 22, which has now been proven false. Others have claimed that two mysterious models, lithiumflow and orionmist, that appeared on the crowd-sourced AI model ranking platform LMArena, could be the company's attempt of testing the model in stealth.

Several X (formerly known as Twitter) users have shared the capabilities of these rumoured models, claiming improved performance in reasoning and front-end coding. Others have claimed it will be a trillion-parameter mixture of experts (MoE) architecture with 15-20B tokens active per query with features such as real-time 60fps video processing, in-built Deep Think reasoning, and a response time of less than one second (with the Flash variant).

Despite the confirmation, Google has not revealed when exactly might this AI model be released to public. Some rumours claim that it could be released between November 15 to December 5, but these are mere speculations. Additionally, these models are likely to be rolled out to a small user group in the US, before expanding it to global user base, enterprise users, and via application programming interfaces (APIs).

Notably, with the Gemini 2.5 models, the tech giant unified the general-purpose and the reasoning versions of the models, by adding native reasoning. Gemini 2.5 Pro was also claimed to outperform leading models of the time such as OpenAI's o3-mini, Grok 3 Beta, Claude 3.7 Sonnet, and DeepSeek R1 in several benchmarks, such as GPQA Diamond, AIME 2024 and 2025, Aider Polyglot, and MMMU.