Grok 3 impresses – But the first reactions suggest that Openai remains to come

• The XAI of Elon Musk deployed Grok 3, saying that it surpasses the OPENAI GPT-4O in key references.
• Grok 3 introduced Deepsearch, a tool to improve research and reasoning.
• Some AI experts have said that Grok 3 is a solid competitor but remains behind Openai in key areas.

Elon Musk presented Grok 3 on Monday, the latest version of Xai’s chatbot, calling him “an order of magnitude more capable” than his predecessor.

Some IA researchers and experts put Grok 3 during early tests – and their initial reactions have been mixed.

Musk said that the new model exceeds GPT-4O of Openai and the Google, Deepseek and Anthropic competitors on key references in mathematics, science and coding.

“We are continuously improving the models every day, and literally within 24 hours, you will see improvements,” Musk said during a live presentation on his XA social media platform alongside three XAI engineers.

Andrej Karpathy, the computer scientist who co -founded Openai and previously directed the operations of Tesla AI, said in a post Monday X that Grok 3 has succeeded in tasks such as the search for complex information from a Training document. However, like O1-Pro of Openai, he had trouble solving “delicate” tick-toe boards, he said.

“The whole impression I got here is that it is somewhere around the O1-Pro capacity, and ahead of Deepseek-R1, although we of course need real and real assessments To examine, “added Karpathy, who had early access to Grok 3.

XAI has also introduced a new tool for Grok 3 called Deepsearch, which aims to improve research, brainstorming and data analysis while explaining how it treats requests and formulates the responses. The startup had worked on the tool Since at least at the end of last year, Business Insider has reported for the first time.

Karpathy said Deepsearch is “approximately around the Depth Research Offer of Perplexity, which is great, but not yet in terms of the recently published” deep research “of Openai, which is always more in -depth and reliable” .

Ethan Mollick, A teacher at Wharton school of the University of Pennsylvania who does a research on AI, Said that, although XAI has quickly set to scale, Grok 3 is not a clear leader.

“A very solid border model, but not the one for which you would stop using your current frontier model,” wrote Mollick in a Post Linkedin on Monday.

He added that although Grok 3 surpasses some OpenAi models in selected landmarks, he clearly does not exceed O3 of Openai.

Gary Marcus, the founder of geometric intelligence and long -standing criticism of the IA media, was more skeptical.

“Elon Musk promised that Grok 3 would be the most intelligent AI of all time. Spoiler Alert: that was not the case,” wrote Marcus on Subk.

He described the “carbon copy” launch event of previous demos, adding that although Grok 3 shows the progression, there was “no game changer” in his performance compared to the OpenAi models.

Karpathy was more positive, calling for the reasoning capacities of advanced Grok 3 “and highlighting its ability to solve complex coding problems.

Related stories

Musk against Altman

The launch of Grok 3 occurs in the middle of increased tensions between Musk and Sam Altman, the CEO of Openai.

Their rivalry intensified this month when Musk led an unsolicited offer of $ 97.4 billion to take control of Openai, the company he co-founded with Altman in 2015 before moving away three years later due to differences. Last week, the Board of Directors of Openai rejected Musk’s offer.

Mollick said that although GROK 3 is a solid model, OPENAI maintains a strong advantage in industry partnerships and the adoption of businesses.

“I don’t know if Grok has a big opening,” he wrote.

Marcus said that Optai did not yet have to worry.

“Sam Altman can breathe easily for the moment. No major jump forward here,” he wrote, adding that the domination of Openai in access to the calculation and the search scale remains a major gap.

Musk said that Grok 3 has “more than 10 times” the computing power of his predecessor and completed pre-training in early January.

Grok 3 immediately takes place at X’s Premium +subscribers, with XAI which also launches a new subscription service called Supergrok for users accessing the bot via mobile or grok.com.

Musk said that XAI planned to open the previous versions of his models once the last reaches maturity, believing that Grok 3 will reach this threshold “in a few months”.