Google launches its "most intelligent model" Gemini 3
19 Nov 2025, 11:23 AMGemini 3 shows reduced sycophancy, increased resistance to prompt injections and improved protection against misuse via cyberattacks
Team Head&Tale
Google has launched its new version of its artificial intelligence (AI) model Gemini 3, which its CEO Sundar Pichai described it as "our most intelligent model," in a company blog post.
Gemini 3 is the most secure model yet, the blog post noted.
"The model shows reduced sycophancy, increased resistance to prompt injections and improved protection against misuse via cyberattacks," it added.
It also said that Gemini 3 pushes the frontier of multimodal reasoning.
The blog post explained that the new model marked the highest score of 37.5% on the Humanity's Last Exam benchmark, which capture academic reasoning. In comparison its earlier version Gemini 2.5 Pro had scored 21.6% and Anthropic's Claude Sonnet 4.5 had scored 13.7% and OpenAI's GPT.1 had scored 26.5%.
Google is not the only company releasing its updated version of its AI model this year.
Just last week, OpenAI launched ChatGPT 5.1 just a few months after rolling out ChatGPT 5 earlier in August. In September, Anthropic launched Claude Sonnet 4.5 while xAI.
Interestingly, the launch of Gemini collided with the roll out of Grok 4.1 by Elon Musk's xAI, which had launched Grok 4 in July this year.
The launch of the updated versions of their AI models within just months of the roll out of their previous models indicates the pressure the AI companies are to upgrade their features to stay ahead of the competition and the need to stay relevant.



