Elon Musk’s artificial intelligence company XAI officially announced the long -awaited flagship model Gok 3. The new model was available on Monday night with advanced skills added to Gok applications for iOS and web. Normally, Gok 3 will be introduced in the second half of 2024. However, it apparently applies to the difficulty of Musk’s difficulties in the artificial intelligence company.

Much more talented than grok 2

Gok is a model family that Musk developed as OpenAI’s GPT-4O and Google’s Gemini as a competitor. Gok, which has the ability to analyze the image and answer questions, also supports the various features of Musk’s social media platform X. Gok 3 finally released after the months of development, and promises great innovations compared to previous versions.

Musk says that Gok 3 was trained with “ten times more calculation power” than his predecessor Gok 2. According to his predecessor, the gigantic Colossus data center found in Memphis for the training of the model, which was stated to be “much more talented, was used. Approximately 200,000 GPUs are reported in Colossus. In addition, Gok 3’s training database contains expanded data sets such as court files.

Musk, in a statement at the launch Gok 3 at a maximum level “a reality seeking artificial intelligence” describing the model sometimes contradicts the political righteousness, even if the truth will say the truth, he said.

Different Gok 3 versions

XAI says Gok 3 is not just a single model, but a model family with different versions. The XAI has not yet shared all the model details on the website. Gok 3 mini compromises a little bit of accuracy when giving faster responses. Gok 3 Reasoning and Gok 3 Mini Reasoning offer deeper thinking about mathematics, science and programming. In this area, where artificial intelligence, such as OpenAI’s O3-Mini model and China-based Deepseek’s R1 model, compete, XAI’s claim is great.

The XAI suggests that Grok 3 has left the GPT-4O behind in certain tests. In particular, a new mathematics named AIME 2025 is claimed to be more successful than competitors.

Users will be able to experience Gok 3 through Gok application. Particularly complex questions, “Big Brain” mode is presented and this mode is able to analyze deeply using extra calculation power.

Deeper research with Deepsearch

In the meantime, Musk says that they are trying to prevent “distillation” activities used to educate another model from a model to protect Gok and XAI’s capabilities. For this reason, some “thoughts ında are hidden in the reasoning processes of the model.

One of the important innovations of Gok 3 is a new feature called Deepsearch. This feature scans the Internet and X platform and offers detailed analysis of users. Deepsearch, which is shown as a rival to OpenAI’s “Deep Research” vehicle, aims to help users more deeply examine complex issues.

How to use grok 3?

Unfortunately, a paid subscription is required to use Gok 3. X Premium+ Subscribers ($ 22/Month) provide priority access to Gok 3, while the new Supergrok plan (probably $ 30/month or $ 300/year) will offer additional advantages such as more deep queries and unlimited images.

In the near future, Grok application will be added to the audio mode. With this feature, Gok will have the ability to speak similar to OpenAI’s Chatgpt. In addition, Gok 3 models will be integrated into XAI’s corporate API in a few weeks.

Musk said that XAI plans to make Gok 2 open source in the coming months: “Our general approach is that when the next version is completely released, we will make the previous Gok version open source. When Gok 3 becomes mature and stable, which will probably be in a few months, then we will make Gok 2 open source. ”