Finally, 'Gemini 3' is here, it's free to use and outperforms GPT-5.1 and Grok 4.1



Google released its AI model ' Gemini 3 ' on Tuesday, November 18, 2025. Gemini 3 has outperformed competing models such as GPT-5.1 and Grok 4.1 in various benchmarks and is already available for free.

Gemini 3: News and announcements

https://blog.google/products/gemini/gemini-3-collection/

Gemini 3: Introducing the latest Gemini AI model from Google
https://blog.google/products/gemini/gemini-3/#note-from-ceo

Google has released the Gemini 3 Pro as the first product in the Gemini 3 series. Below is a table listing the benchmark results for Gemini 3 Pro, Gemini 2.5 Pro, Claude Sonnet 4.5, and GPT-5.1. Gemini 3 Pro achieved the highest scores in almost all tests.



Gemini 3 Pro is also participating in

LMArena , where humans evaluate the performance of anonymous AIs, and in the Text Arena, which competes for text generation performance, it has surpassed Grok 4.1 , which was released on November 18, 2025, to take first place.



It also recorded scores in

the Vision Arena , a competition to test image comprehension, far exceeding those of OpenAI's model and Google's previous model.



It also came in first in

web development tasks, beating out OpenAI and Anthropic models.



In addition, Gemini 3 Deep Think has also been developed to handle complex tasks. The following figure compares the benchmark results of Gemini 3 Deep Think, Gemini 3 Pro, Gemini 2.5 Pro, Claude Sonnet 4.5, GPT-5 Pro, and GPT-5.1, showing that Gemini 3 Pro scores higher than other models and Gemini 3 Deep Think achieves the highest score.



The results of

the ARC-AGI-2 benchmark test, developed to measure AI intelligence, are below. Gemini 3 Pro achieved a higher score than GPT-5 Pro at a lower cost. Gemini 3 Deep Think, while more expensive than other models, achieved an overwhelmingly higher score.



Gemini 3 Pro is already available on the Gemini web app and smartphone app, and it's available for free users, so I tried it out.

First, access

the web app version of Gemini in your browser, click the mode selection button at the bottom right of the input field, and then click 'Thinking Mode.'



Click Thinking Mode, enter a question, and Gemini 3 Pro will respond. According to Google, Gemini 3 is powerful enough to understand the context and intent behind the user's instructions and respond accordingly. In this case, I entered a vague question, and even made a typo, writing 'ochi' (fall) as 'okiochi' (place fall), but the Gemini 3 Pro carefully interpreted it and responded.



Click 'Show thought process' to see the thought process leading up to the answer.



However, at the time of writing, the thoughts were displayed in English.



Gemini 3 is also integrated into Google Search's AI mode. However, at the time of writing, this is only available in the US, and in Japan, only the traditional AI mode is available.




Gemini 3 Deep Think will be released to Google AI Ultra subscribers in the coming weeks after safety testing and other procedures are completed. Other models in the Gemini 3 series are also planned for release in the near future.

Related Posts:

in AI, Posted by log1o_hf