Last week, Elon Musk made a major announcement for the AI industry. Grok, the flagship product from xAI, would be getting its Grok 3 release in a couple of weeks or two. As promised, Musk’s AI-focused company has begun rolling out Grok 3.
Grok 3 brings a series of models that focus on different types of reasoning
Grok 3 takes a similar approach to Gemini or ChatGPT by having a series of models for different goals. The new release seeks to compete directly against other top AI-powered models like OpenAI’s GPT-4, Gemini, and newcomer DeepSeek.
“Grok 3 is an order of magnitude more capable than Grok 2,” Musk said. More specifically, xAI used a massive data center with around 200,000 GPUs to train Grok 3. It required “10x” more computational power than its predecessor, the X owner claims. xAI promises that its AI models are “maximally truth-seeking AI, even if that truth is sometimes at odds with what is politically correct.”
xAI boasts about the power of Grok 3, claiming that it outperforms GPT-4o on benchmarks like AIME (focused on math) and GPQA (focused on PhD-level physics, biology, and chemistry applications). However, it’s not all about complex reasoning in xAI’s latest AI. There is Grok 3 mini, a model focused on speed, although it sacrifices some accuracy in the process. It’s ideal for everyday requirements, similar to the Gemini Flash models from Google.
Grok 3 Reasoning and Grok 3 mini Reasoning, response to GPT o3-mini
xAI also announced Grok 3 Reasoning and Grok 3 mini Reasoning, models that appear to be direct responses to GPT o3-mini and o3-mini-high. The firm boasts that, once again, its artificial intelligence emerged victorious in key benchmarks. The list includes AIME 2025, the latest version of the AI math benchmark. Grok 3 Reasoning models can “think through” problems, which takes longer in favor of more accurate responses to complex requests.
Humans can trigger Grok’s complex reasoning capabilities using the “Big Brain Mode.” xAI recommends using the option for requests related to math, science, and coding.
Recently, OpenAI and Microsoft began investigating whether DeepSeek distilled data from GPT models to train R1 and V3. To avoid this, Grok 3 “obscures” some thoughts in its reasoning chain. Grok 3 also offers a response to ChatGPT’s DeepSearch feature. That means you can “assign” the AI to do research on a specific topic. This feature is especially useful for complex tasks.
Elon Musk’s AI to get a voice soon
There are options still in development that will be coming soon. One of them, a “voice mode,” is arriving next week, Musk says. This will finally give Grok a synthesized voice, something it still lacks. Later on—no exact date yet—xAI will launch the enterprise API with support for DeepSearch so others can build services and apps on top of it.
Grok 3 is reaching first X’s Premium+ tier subscribers who pay $50/month. According to leaks, a “SuperGrok” plan will be coming later that could cost $30/month or $300/year and offer all the full reasoning power of Musk’s AI, among other perks.
Grok 2 becoming open-source soon
Lastly, Grok 2 will become open-source in the coming months, as confirmed by Musk. “Our general approach is that we will open-source the last version [of Grok] when the next version is fully out,” he continued. “When Grok 3 is mature and stable, which is probably within a few months, then we’ll open-source Grok 2,” he said.
