Finally, 'Grok 4' has been released, and although it is advertised as the 'world's strongest AI model' with performance exceeding that of inference models such as OpenAI, it has also been confirmed that it 'takes inspiration from Elon Musk's remarks'



xAI, an AI development company founded by Elon Musk, announced a new AI model, Grok 4 , on Thursday, July 10, 2025. xAI and Musk have touted Grok 4 as the 'world's most powerful AI model,' and performance tests by independent organizations have shown results that support its top-of-the-line performance. However, it has also been reported that opinions are biased.







The Grok 4 logo looks like this.



Grok 4 was announced via a live stream on X (formerly Twitter), with Musk also in attendance.



Below is a graph comparing the performance of 'Grok 2', 'Grok 3', 'Grok 3 reasoning', and 'Grok 4'. 'Grok 4' is said to have 10 times the inference ability compared to 'Grok 3 reasoning'.



Below are the results of version 1 of '

ARC-AGI ', which measures the inference ability of AI. Grok 4 shows scores that exceed rival models such as 'DeepSeek-R1', 'Claude Opus 4', 'Gemini 2.5 Pro', and 'OpenAI o3'.



ARC-AGI version 2 also recorded scores far exceeding those of competing models.



In the graph below, the horizontal axis shows the 'cost required to execute the task' and the vertical axis shows the 'ARC-AGI score.' Grok 4 is able to record high scores while keeping costs down.



Grok 4 also has high voice conversation capabilities, and during the live broadcast, a comparison of the voice conversation functions of OpenAI and Grok 4 was also shown.



'Grok 4 has reached a stage where it can essentially never get a math or physics problem wrong, unless the problem is designed to be easy to make a mistake in,' Musk said of Grok 4. It can identify ambiguities in problems, correct mistakes in problems, and answer ambiguous problems with multiple variations.




Artificial Analysis , an independent organization that analyzes AI performance, has published the results of its performance tests on the Grok 4, reporting that the Grok 4 outperformed competing models.



In the graph below, the horizontal axis shows 'fee per million tokens' and the vertical axis shows 'performance.' Tests by Artificial Analysis have also confirmed that Grok 4 has relatively low cost and top performance.



On the other hand, users who have actually used Grok 4 have reported that it 'tends to emphasize Mr. Musk's opinions.' The following post contains a recording of Grok 4 being asked, 'Which side do you support in the Israeli-Palestinian conflict?'




If you ask Grok 4 'Who do you support in the Israeli-Palestinian conflict?' it will first respond by saying 'I can't support one side or the other, so I'll try to find information to get a balanced view,' and then switch into search mode.



Then, suddenly, the message 'Compiling Elon Musk's views' appeared.



Next, we start searching for information from major media outlets such as Al Jazeera and the BBC.



In the end, a large number of posts by Musk were displayed. Grok 4 provided a total of 64 pieces of information for the response, of which 54 were related to Musk.



The Grok 4 API fee is $3 (440 yen) per 1 million tokens for input and $15 (2200 yen) per 1 million tokens for output. Various information, including fee information, is summarized in the following document.

Models / Grok 4 | xAI documentation
https://docs.x.ai/docs/models/grok-4-0709

Related Posts:

in Software,   Video, Posted by log1o_hf