Elon Musk’s Ai, Xai, released late on Monday, released the latest pioneering AI, Grok 3, and revealed new capabilities for Grok iOS and web applications.
Grok, Xai’s answer to models such as Openai’s GPT-4O and Google’s Gemini can analyze images and answer questions, and ventilate a number of features on the social network of Musk, X. GROK 3, which was under development for several months, was optimistic. It is scheduled to be launched in 2024, but he was absent from the deadline.
The two is the launch of an ambition.
Xai uses a huge data center in Memphis that contains about 200,000 graphics processing units for GROK 3. mail On X, Musk claimed that Grok 3 was developed using “10x” computing power more than its predecessor, GROK 2, using an extensive training data set that includes outwardly Files from court cases.
“Grok 3 is more than Grok 2,” Musk said during a direct presentation on Monday. “[It’s a] To the maximum extent of artificial intelligence, even if this fact is sometimes contradicts what is correctly true. “
Grok 3 is a family of models, for accuracy. A smaller version of GROK 3, GROK 3 MinI, is more quickly at the expense of some accuracy. Not all the relevant models and features are available to GROK 3 yet (some in the experimental version), but they started to put it on Monday.
Xai Grok 3 Beats GPT-4O claims the standards including AIME (which evaluates the performance of the model in taking samples of mathematics questions) and GPQA (which evaluates models using physics at the level of PhD, biology, and chemistry problems). An early version of Grok 3 also recorded competitively in Chatbot Arena, a group test that stimulates different AI models against each other and users vote on their favorite responses, according to Xai.
Models in the new Grok 3 family, GROK 3 Reasoning and Grok 3 Mini Reasoning, can “think through” carefully, similar to “thinking” models such as Openai’s O3-MINI and the Chinese company Deepseek’s R1. Thinking models try to check the facts before giving the results, helping them avoid some of the pitfalls that usually rise.
XAI claims that Grok 3 Reasoning exceeds the best version of O3-MINI-O3-MINI-HIGH-in many popular standards, including the newest mathematics standard called AIME 2025.
Thinking models can be accessed via the GROK application. Users can ask Grok 3 “Thinking”, or – for the most difficult information – take advantage of the “Big Brain” mode for thinking that uses additional computing. Xai describes thinking models as the most appropriate for mathematics, science and questions for programming.
Musk said that some “ideas” thinking models are blocked in the Grok application to prevent distillation, a method used by the developers of the artificial intelligence model to extract knowledge from other models. Recently, Deepseek has been accused of distillation of Openai models for its own creation.
Grok thinking models support a new feature in the GROK application called Deepsearch, XAI’s answer to the Acting Search tools such as deep search in Openai. Deepsearch scans Internet and X to analyze information and provide a summary in response to a question.
The participants in X Premium+ Tier ($ 50 a month) will be able to reach Grok 3 First, and other features will be placed behind a new plan that calls it to Xai Supergrok. At a price of $ 30 per month or $ 300 a year (If you are supposed to believe the leaks), Supergrok opens additional inquiries and inquiries, and throws unlimited images.
In the future – just about a week from now – the GROK app will get a “audio mode”, said Musk, which will give Grok models made. A few weeks later, Grok 3 models will be available via Enterprise, along with Deepsearch.
Musk said that XAI planned the source of the source Grok 2 in the coming months.
“Our public approach is that we will open the source the last version [of Grok] When the next version is completely.
When Musk Grok announced almost two years ago, he developed the artificial intelligence model as an exciting, non-candidate, and anti-“waking”-in general, ready to answer the controversial questions that other artificial intelligence systems will not do. He greeted some of this promise. It is required to be vulgar, for example, that GROK and Grok 2 will adhere to happiness, which puts a colorful language that you may not hear from ChatGPT.
But Grok Models before Grok 3 Hedge On political issues and will not cross Certain limits. In reality, One study I found that Grook bent over the political left on topics such as transgender people, diversity programs and inequality.
Musk blames behavior for GROK training data – public web pages – and pledge To “convert Grok closer to political neutral”. It is not yet clear whether Xai has achieved this goal, and what are the consequences.
