Elon Musk’s AI Company, Xai releases its latest flagship model, the Grok 3

Xai, the AI ​​company of Elon Musk, released its latest flagship AI model, the Grok 3 late Monday night, releasing new features in the Grok app for iOS and the web.

GlockAnswer to a model like Xai’s Openai GPT-4O And Google’s GeminiIt can analyze images, respond to questions, and promote many features of the X.Grok 3, Musk’s social network that has been developed for several months. It is scheduled with optimism It was released in 2024, but missed the deadline.

Monday is an ambitious launch.

Xai uses a huge data center in Memphis (a data center containing about 200,000 GPUs) to train the GROK 3. post In X, Musk argued that the Grok 3 was developed with “10 times” computing than its predecessor, the Grok 2, and, on the surface, was developed with the expanded training dataset included. Filing from lawsuit.

“The Grok 3 has more single digit capabilities than the Grok 2,” Musk said in a live streaming presentation Monday. “(It’s) an AI seeking the most truth, even if it sometimes conflicts with what is politically correct.”

The Grok 3 is a family of models, not just one, to be exact, but a family of models. The Grok 3 Mini, a small version of the Grok 3, responds quickly to questions at the expense of some degree of accuracy. Not all models are available yet (and some are in beta), but the rollout starts on Monday.

Xai claims that the Grok 3 defeated GPT-4O on benchmarks that include AIIME. This evaluates the performance of the model regarding sampling mathematics questions and evaluates the GPQA, which uses PHD-level physics, biology, and chemistry problems to evaluate the model. The early version of GROK 3 also recorded a competitive score Chatbot ArenaAccording to Xai, a crowdsourcing test in which different AI models pit each other and vote for user preference responses.

Two variations of Grok 3, Grok 3 Reasoning and Grok 3 Mini Reasoning, like “inference” models like Openai, allow you to carefully “consider” the problem. o3-mini R1 from Chinese AI company Deepseek. The inference model should thoroughly check yourself before producing results. Avoid some of the pitfalls It usually trips the model.

Xai claims that Grok 3 reasoning surpasses the best version of O3-Mini (O3-Mini-High) in several popular benchmarks, including a new mathematical benchmark called AIME 2025.

The inference model can be accessed via the GROK app. Users can ask Grok 3 to “think.” Alternatively, for more difficult queries, you can leverage the “big brain” mode for inference using additional computing. Xai describes inference models as the best questions for mathematics, science and programming related questions.

According to Musk, in the Grok app, some of the “thinking” in the inference model is obscured to prevent distillation, a method used by AI model developers to extract knowledge from another model. He said. Recently, Deepseek was like that Accused of distilling Openai’s model Create your own.

Grok’s inference model supports a new feature in the Grok app called deepsearch. In-depth study of Openai. DeepSearch scans the Internet and X to analyze information in response to questions and provide abstraction.

Subscribers of X’s Premium + Tier You first get the Grok 3 and other features are gated behind the new plan called Xai’s SuperGrok. SuperGrok unlocks additional inference and DeepSearch queries and throws unlimited image generation at a price of $30 per month or $300 per year.

In the future, Musk said the Grok app will acquire “voice mode” about a week from now. A few weeks later, the Grok 3 model will arrive Xai’s Enterprise APIalong with the DeepSearch feature. A few months later, Musk said Xai will make Grok 2 open source.

“Our general approach is to open source the last version (of Grok) when the next version is completely out,” Musk says. “If Grok 3 is mature and stable, and if it’s probably within a few months, I’ll make Grok 2 open source 2.”

When Musk unveiled Grok about two years ago, he pitched AI as an edgy, filtered, anti-‘awakening’. He told me some of those promises. For example, Grok and Grok 2, who are said to be vulgar, are willing to obligate and spit out colorful languages ​​you haven’t heard chatgpt.

However, GROK models before Grok 3 hedge There is no intersecting about political subjects Specific boundaries. in fact, One study Grok found that he left on topics such as trans rights, diversity programs, and inequality.

Musk criticizes the actions of Grok’s training data (public web pages); Pledge “Take Grok closer to political neutrality.” It is not yet clear whether Xai has achieved that goal.