Elon Musk’s Ai, Xai, Final Monday has issued its latest AI, Grok 3 model, and new Grok ios and web systems.
Grok answer, models similar to Opelai-4O’s Gemini, can analyze questions, and provide many features in Musk’s Social Network, X. Progress-advanced, and missed that limit.
Monday is the beginning of prominence.
XAi has been using the main center of the Memphis Data Containing approximately 200,000 GPUS training Grok the filling of court cases.
“Grok 3 is a great performance in size than Grok 2,” Musk said during a live renewed presentation on Monday. “[It’s a] The truth seeking true truth, even if that fact sometimes conflicts with the right political. “
Grok 3 is a model family, to clarify. Grok 3, Grok 3 Mini, answers questions as soon as possible for some accuracy costs. Not all models and features related to Grok 3 are available but (some Beta), but began to go out on Monday.
The Grok 3 Beats GPT-4O in benches include AIME (checking the functioning of the model questions) and the GPQA (Models using Physics, Bioology, and Chemistry problems). The original Greek type also looks at the mind and Chatbot Arena, full test groups including different AI models and users vote for their selected responses, according to Xai.
Two variations of the new model, Grok 3 Reasoning and 5 mini thoughts, can carefully “think” of problems, such as Rupels’s O3-Mini Company Depseek’s R1. Reasonable models try to re-check before providing results, which helps them avoid some of the issues that often move models.
XAi says Grok 3 Reasoning exceeds the best type of O3-Mini – O3-Mini-High – in several popular benchmarks, including a new benchmark of AIMPICA 2025.
These types of consultation can be obtained by the Grok app. Users can ask Grok 3 to “think,” or – to get the hardest questions – the profit “Big Brain” of the consultation mode using additional computing. Xai defines imaginary models are very good for statistics, science, and programming questions.
Musk said some models of models said that “thoughts” are hidden in the Grok app to prevent the installation, the method used by AI modeling information from other models. Recently, Deepseek is accused of Openaai models that sprouts to create their own.
Greek consulting models reduced the new feature to the Grok app called Deepsearch, the response of the research tools that are enabled as a deep research for Opelai. Deepsearch scaning Internet and X analyzes information and brings uncomfortable in answering the question.
Subscribers in X’s Premium + Tier ($ 22 per month) will receive access to Grok 3 first, and some features will be closed after a new new cost plangrok. Price for $ 30 a month or $ 300 per year
Next time – nearly a week from now – the Grok app will receive a “voice mode,” Musk said, which will provide the Groker models by a tone. A few weeks thereafter, Grok 3 models will be available in business API, as well as Deepsearch power.
Xai plans to open – Source Grok 2 in the coming months, Musk said.
“Our usual method is to open – a source of last version [of Grok] When the next version is completely over, “continued.” When Grok 3 is mature and stable, likely within a few months, then we will open – Source Grok 2. “
When the Musk announces Grok two years ago, he set up ai model as Edby, Unused and Fighted, in a determination, answering the contrary questions Some Ai programs will not. He brought another promise. He was told to be obscene, for example, Greek and Grok 2 would drop in a wholehearted, and postponed a colorful language, who may not hear you in Chatgt.
But Grok models before Grok 3 in the footsteps of politics and did not reject certain boundaries. In fact, one lesson found that Grok is dependent on politics such as transgender rights, diversity, and inequalities.
Musk also blamed the behavior of Grok’s training – a Web community pages – and are affordable to the “Shift Grok near Neutrality.” It is not clear whether Xichai earned that purpose, and what is bad consequences.
Source link
