Elon Musk’s AI company, xAI, has released Grok 3 alongside other new features in the Grok apps for iOS and the web. The chatbot, released on Monday, is the latest flagship AI model in the Grok world.
Grok, just like OpenAI’s GPT-4o and Google’s Gemini, can analyze images, respond to questions, and power several features on Musk’s social network, X.
The latest model of Grok, Grok 3, has been in development for quite a while. The model was previously stated to be released in 2024 but missed the deadline.
Elon Musk described the model via an X post, claiming that Grok 3 was developed with “10x” more computing power than its predecessor, Grok 2. It also comes with an expanded training data set that includes filings from court cases.
During the development stage, xAI has been using an enormous data centre in Memphis (containing around 200,000 GPUs) to train Grok 3.
“Grok 3 is an order of magnitude more capable than Grok 2. It’s a maximally truth-seeking AI, even if that truth is sometimes at odds with what is politically correct,” Musk said during a live-streamed presentation Monday.
Grok 3 comes with a bunch of models with the rollout being deployed phase by phase. For instance, a smaller version of Grok 3, Grok 3 mini, responds to questions more quickly at the cost of some accuracy.
Also Read: South Korea pauses DeepSeek’s download amid privacy concerns.
Grok 3’s reasoning model
The AI company asserts that the latest model surpassed GPT-4o on measures such as AIME and GPQA. The former evaluates a model’s performance on a sampling of math questions, while the latter assesses models using PhD-level physics, biology, and chemistry problems.
According to xAI, an early version of Grok 3 scored competitively in Chatbot Arena – a crowdsourced test that pits different AI models against each other and has users vote on their preferred responses. It added that the model surpasses the best version of the o3-mini on several popular benchmarks, including a newer mathematics benchmark called AIME 2025.
In the Grok 3 circle, models such as Grok 3 Reasoning and Grok 3 mini-reasoning can carefully “think through” problems. xAI claims this feature is similar to “reasoning” models like OpenAI’s o3-mini and Chinese AI company DeepSeek’s R1.
xAI describes the reasoning models best suited for mathematics, science, and programming-related questions. While the reasoning models can be accessed via the Grok app, users can ask Grok 3 to “Think,” or leverage “Big Brain” mode for reasoning that employs additional computing and for more difficult queries.
On the training mechanism behind Grok’s capability, Musk said that some of the reasoning models’ “thoughts” were hidden to prevent distillation, a method used by AI model developers to extract knowledge from another model.
Recall that Chinese DeepSeek was recently accused of distilling OpenAI’s model to create its own.
xAI noted that subscribers to X’s Premium+ tier ($22 per month) will get Grok 3 first, while other upcoming features will be ushered in by a new plan called SuperGrok. The model, projected to price at $30 per month or $300 per year, will unlock additional reasoning, and DeepSearch queries, and provide unlimited image generation.
Going forward
According to Elon Musk, the Grok app will gain a “voice mode,” which will give the models a synthesized voice. Later on, Grok 3 models will arrive in xAI’s enterprise API, along with the DeepSearch capability. The process and launch are slated for a month frame.
Also, xAI plans to open-source Grok 2 in the coming months.
“Our general approach is that we will open-source the last version [of Grok] when the next version is fully out. When Grok 3 is mature and stable, which is probably within a few months, then we’ll open-source Grok 2,” Musk noted.
At the launch of Grok two years ago, Musk claimed that the model was superior and would answer controversial questions other AI systems wouldn’t, though some of those promises were fulfilled.
A delimiting factor in the Grok model is that when prompted on political subjects, it leaned to the political left on topics like transgender rights, diversity programs, and inequality thereby negating fairness. Musk blamed this behaviour on Grok’s training data and pledged to “shift Grok closer to politically neutral.”
With a new launch, Grok’s reasoning models underpin a new feature in the Grok app called DeepSearch, which is xAI’s answer to AI-powered “deep research” tools like OpenAI’s deep research. DeepSearch scans the Internet and X to analyze information and deliver an abstract in response to a question.
Technext Newsletter
Get the best of Africa’s daily tech to your inbox – first thing every morning.
Join the community now!
