Elon Musk’s XAI announced the latest flagship AI model, the Grok 3. The company also announced the Grok 3 Mini, a zoomed version, plus the addition of DeepSearch, a new tool the company calls the next-generation search engine. XAI adds new features for Grok 3 web and mobile applications, as well as a subscription service specifically for Grok users, called SuperGrok.
“We’re excited to introduce the Grok 3, which is an order of magnitude more capable than the Grok 2 in a very short time,” Musk said in a live broadcast on Xai on X. The H100 GPU is twice as good as the Grok 2. The team said it took 92 days to expand its Memphis supercomputer, known as Colossus, to accommodate the training of the Grok 3.
Musk said in his speech that the Grok 3 has 15 times more computer power than the Grok 2, although in previous X posts, he said 10 times more. The model is trained in information from user posts on X to court documents.
Grok 3 faces fierce competition from Openai, Google and Anthropic, all of which have released new AI models in 2025, or are planning to do so.
Google Gemini 2.0 added useful features in early February, while OpenAI plans to unify all its AI models when it launches GPT-5 later in 2025. Meanwhile, Anthropic’s next new AI model could be weeks away.
How to try Grok 3
Grok 3 will be launched to X Premium Plus members starting February 18, and they will now have exclusive access, including access to DeepSearch. This is a small win for these users, as X recently raised the price of the Premium Plus from $16 to $22.
Eventually, XAI will launch SuperGrok, a subscription service specifically for Grok 3, which includes DeepSearch, higher image generation restrictions and access to Grok 3 Mini features (such as Think). Prices were not displayed during the demonstration.
The benchmark compares the Grok 3 and Grok 3 Mini to the match.
“Scary clever”
“Grok 3 has a very strong reasoning ability,” Musk said in an interview with CNBC on February 13. “In the tests we have done so far, the Grok 3 outperforms anything we know. Sometimes , I think the Grok 3 is horrible. ”
Of course, there is no feature to test Grok 3 yet. The team mentioned that Groke might answer believe It’s true – otherwise there is no mention of the Grok 2’s preference output hallucination or how often Grok 3 would do so.
Grok 3 shows an in-depth search on SpaceX’s next Starship release.
DeepSearch
In addition to Grok 3, the XAI team has also announced DeepSearch, described as the first generation of Grok 3 agents, allowing users to ask questions and receive answers. It is called the “next generation search engine” in live streaming.
Openai and Google have similar proxy-based searches. Both are called in-depth research.
DeepSearch shows users the experience of a single step 3, from thinking about problems to research, and then answers. The demo took about a minute and included 15 X posts and 32 web pages as references.
Another feature of DeepSearch is the ability to view Grok 3’s reasoning. After asking a question about March Madness, the team used the feature to show how Grok 3 came to the conclusion.
“Maybe, I should look at team rankings, performances in the regular season, any injuries or key player status, and maybe some historical data about their performance in the game,” Grok 3 revealed during an inquiry.
This is after the Chinese artificial intelligence company DeepSeek (which launched the platform in January) – including the inference process while answering queries.
Grok 3 conducted an in-depth search on SpaceX’s next Starship release.
Grok vs. competitors
Grok is not alone, an expert told CNET that there might be something to catch up.
Paul Schell, an industry analyst at ABI Research, said: “While the advanced benchmark results are impressive, it illustrates more about the general movement of models with reasoning capabilities or proxy AI than the impact on the individual competitiveness of the model. “We considered a wider listing, which covered marketing strategies and partnerships built by model builders. Here, Human and Openai are ahead of Grok.”
The difference between Grok and rival Schell Sais is that “Grok 3’s DeepSearch is easy to access real-time information from the Internet, unlike OpenAI’s Deep Research product.”
But he also noted that when searching, Openai’s product “seems to be deeper” and takes up to 30 minutes to complete a task.
