'Grok-3' will be released, and xAI claims it has overwhelmingly higher performance than GPT-4o and Gemini 2.0 Pro



Elon Musk's AI company 'xAI' announced Grok-3 on February 18, 2025. xAI claims that Grok-3 has more than 10 times the computing power of its predecessor, and is overwhelmingly more powerful than GPT-4o and Gemini 2.0 Pro.





Grok-3 has two reasoning modes: 'Think', which displays Grok's reasoning as it resolves requests, and 'Big Brain', for more complex tasks that require more computational power.



Below is the actual question I asked in the Grok app, using the Grok-3's Think mode. The answer was a bit of a roundabout calculation, but it was still a valid proof.



There was no button for Big Brain mode, but according to Grok-3, you can use it by specifying 'answer in big brain mode' in the prompt. At the time of writing, there was no help page, so it is unclear whether Grok-3's advice is correct.



Grok-3 is a multimodal model, so it can load not only text, but also images, PDF files, etc.



xAI has also released 'DeepSearch,' which scans the Internet and X to provide detailed answers to users' questions. When a user enters a question into Grok-3, it first analyzes the intent and keywords of the question, understands what the question is asking, generates an appropriate search query, and the AI model filters it to generate an answer.



DeepSearch also visualizes the inferences made during a search. However, when I actually used DeepSearch to ask a question in Japanese and looked at the inferences, it seemed to be thinking entirely in romaji. Moreover, there were some errors in the pronunciation of kanji, but it seemed to be able to find the desired information on the Internet and X.



There are two models of Grok-3: Grok-3 and Grok-3 mini. In the following post, xAI presents a graph comparing the benchmark results of Grok-3 and Grok-3 mini, Gemini-2 Pro and DeepSeek V3, Claude 3.5 Sonnet, and GPT-4o. According to this, Grok-3 and Grok-3 mini outperformed competitors including Gemini-2 Pro, DeepSeek-V3, Claude 3.5 Sonnet, and GPT-4o in several tests including mathematics ( AIME ), science ( GPQA ), and coding ( LCB ).



'The mission of xAI and Grok is to understand the universe. What's going on? Where are the aliens? What is the meaning of life? How is the universe going to end? How did the universe begin? We want to answer those questions,' Musk said at the beginning of his presentation. 'Of course, Grok will be an AI that seeks the truth as much as possible, even if that truth contradicts what is politically correct.'

At the time of writing, Grok-3 can be used from the Grok app for smartphones. However, in order to use Grok-3, you need to sign up for a paid subscription, 'X Premium Plus' or higher.

Immediately after the release of Grok-3, X raised the price of X Premium Plus in the United States from $22 (approx. 3,340 yen) per month to $40 (approx. 6,000 yen) and from $ 229 (approx. 35,000 yen) per year to $395 (approx. 60,000 yen). At the time of writing in Japan, the monthly fee is 2,590 yen (tax included) and the annual fee is 27,300 yen, which is the same price as in December 2024 .

Related Posts:

in Review,   Mobile,   Software,   Web Service,   Video, Posted by log1i_yk