Grok

Grok is a computer program you can have a written conversation with. You type in a question, and the program writes back an answer in everyday language. It is developed by the company xAI, which Elon Musk founded in 2023. Grok is embedded above all in the social network X, formerly Twitter, which also belongs to Musk. There, you can address the bot directly under posts and ask it for explanations. Its tone stands out: Grok is deliberately designed to be cheekier and funnier than the competition.

Musk’s counter-model to ChatGPT

Grok is Elon Musk’s most important attempt to compete in the race for chat programs. The market leader is called ChatGPT and comes from OpenAI. Musk was once a co-founder and financial backer there himself, but fell out with the company. Grok is therefore also a personal counter-model, not just a product.

Musk’s criticism is that other programs are too politically cautious and refuse to answer too often. Grok, by contrast, is supposed to dodge as little as possible. But that is exactly what makes the bot vulnerable to criticism. On several occasions, Grok has produced statements that were classified as offensive or clearly false. In the summer of 2025, a series of antisemitic responses caused international criticism, and xAI had to make improvements and apologized.

For the debate about artificial intelligence, Grok is therefore an important test case. It shows what happens when a system’s safeguards are deliberately loosened. Authorities in the EU keep a close eye on such incidents, because large platforms there face stricter requirements against disinformation.

Language model plus live access to X

At the core of Grok is what is known as a language model. That is a program that has learned from huge amounts of text which word typically follows next. So it always only predicts the next piece of text, word by word. Because it draws on patterns from billions of sentences in doing so, the answers seem like well-thought-out language. The program itself has no real understanding or opinion.

The special feature of Grok is its connection to X. The bot can access current public posts while it is answering. Other chat programs often only know the state of their training, which can lie months in the past. This gives Grok an advantage on topics that are current that day. The downside is obvious: if nonsense happens to be circulating on X, that nonsense can end up in the answer.

Grok appears in numbered versions, such as Grok 3 or Grok 4. Each new version is trained with more computing power and performs better in tests. Training takes place in xAI’s own data center in Memphis, which the company calls Colossus. Tens of thousands of specialized chips are housed there, handling the calculations in parallel.

From the X app to the Tesla center console

Most often, you encounter Grok in the X app. Paying subscribers can message the bot directly there. Under a post, you can also ask what it’s actually about. Grok is also available as its own app and as a website, in some cases for free with a limited number of queries.

Beyond that, Musk is building the bot into his other companies. In newer Tesla vehicles, Grok can be addressed via the screen in the car. Companies can also purchase the technology through an interface and build it into their own products. For developers, that is the actual business purpose.

In business news, Grok mostly comes up in two contexts. Either it’s about the valuation of xAI, which has since merged with X and is raising billions in investor money. Or it’s about a dispute because the bot has put out something problematic. Incidentally, a common misconception is that Grok is a search engine. It is a text generator, one that can also get things wrong, which is why it is worth double-checking on important questions.