Elon Musk’s AI company, xAI, releases its latest flagship model, Grok 3

The AI ​​company of Elon Musk, Xai, released its latest flagship AI model, Grok 3, late on Monday evening, together with new possibilities in the Grok -apps for iOS and the internet.

Grok, Xai’s answer to models such as OpenAI’s GPT-4O and Google’s Gemini, can analyze and respond to questions, and a number of functions on the social network of Musk, X. Grok 3, which has been in development for several months, was optimistic in development. Planned for release in 2024, but missed that deadline.

Monday is an ambitious launch.

Xai has used a huge data center in Memphis – a data center with around 200,000 GPUs – to train grok 3. In one after On X, Musk claimed that Grok 3 was developed with “10x” more computing than grok 2, its predecessor, and with an extensive training data set that apparently includes Archiving of lawsuits.

“Grok 3 is an order of size that is better suited than Grok 2,” said Musk during a live-ramed presentation on Monday. ‘[It’s a] Maximum seeking AI, even if that truth is sometimes at odds with what is politically correct. “

Grok 3 is a family of models, to be precise – not just one. A smaller version of Grok 3, Grok 3 Mini, responds faster to questions at the expense of any accuracy. There are not all models and related functions available yet (and some are in beta), but the rollout starts on Monday.

Xai claims that Grok beats 3 GPT-4O on benchmarks, including Aime, who evaluates the performance of a model about a sample of mathematics questions, and GPQA, which assesses models with the help of PhD level physics, biology and chemical problems. An early version of Grok 3 also scored competitive in Chatbot Arena, a crowdsourced test that takes different AI models against each other and let users vote for their preferred reactions, according to Xai.

Two variations of grok 3, grok 3 reasoning and grok 3 mini-reasoning, can carefully ‘think’ problems, similar to models ‘reason’ such as OpenAIs O3-Mini and Chinese AI Company Deepseek’s R1. Reasoning models check themselves thoroughly before they give results, which helps them to avoid some of the pitfalls that normally stumble on models.

Xai claims that Grok 3 reasoning the best version of O3-Mini-O3-Mini-high-transfers at various popular benchmarks, including a newer mathematical benchmark called AIME 2025.

The reasoning models are accessible via the Grok app. Users can ask grok 3 to ‘think’, or – for more difficult questions – use the ‘Big Brain’ mode for reasoning that uses extra computing. Xai describes the reasoning models as the most suitable for mathematics, science and programming questions.

Musk said that in the grok -some of the ‘thoughts’ of the reasoning models are hidden to prevent distillation, a method used by AI model developers to extract knowledge from another model. Deepseek was recently accused of distilling the models of OpenAi to create themselves.

The reasoning models of Grok support a new function in the Grok app called DeepSearch, Xai’s answer to AI-driven “Deep Research” tools such as OpenAi’s Deep Research. DeepSearch scans internet and X to analyze information and deliver an abstract in response to a question.

Subscribers of X’s Premium+ Tier ($ 22 per month) first get grok 3, and other functions are closed behind a new plan that is calling Xai Supergrok. Priced at $ 30 per month or $ 300 per year (Should be believed if leaks), SuperGrok unlocks additional reasoning and deep searchqueries and throws unlimited image generation.

In the future – as soon as about a week from now on – the grok – app will get a ‘speech mode’, said Musk, who gives grok models a synthesized voice. A few weeks later, Grok 3 models will arrive in Xai’s Enterprise API, together with the DeepSearch capacity.

Xai is planning to open grok 2 in the coming months, said Musk.

“Our general approach is that we open the latest version [of Grok] When the next version is completely off, “he continued. “When grok is 3 mature and stable, which is probably within a few months, then we open grok 2.”

When Musk Grok announced about two years ago, he threw the AI ​​as edgy, unfiltered and anti-“wake up”-in general willing to answer controversial questions that other AI systems will not do. He has realized part of that promise. Felled to be vulgar, for example, Grok and Grok 2 would like to oblige, which spits a colorful language that you would probably not hear from chatgpt.

But Grok -Models prior to Grok 3 roofed on political topics and would not cross Certain limits. In reality, One study Discovered that Grok leaned to political links on topics such as transgender rights, diversity programs and inequality.

Musk has blamed the behavior of the training data of Grok – Public web pages – and promised To “move closer to politically neutral.” It is not yet clear whether Xai has achieved that goal – and what the consequences could be.