Elon Musk’s xAI has launched Grok 3, the latest update to its AI model designed to compete with those of OpenAI, Google and DeepSeek. While competition will grow as independent benchmark tests are carried out, Grok 3 has the potential to establish itself as a strong player, explains Jatin Grover
What is Grok 3?
Grok 3 is the latest artificial intelligence model developed by tech billionaire Elon Musk’s xAI. It is a significant update to its predecessors, offering improved performance in tasks like mathematics, science, and coding. Grok 3 introduces features such as Think Mode, which provides step-by-step reasoning for users, and Big Brain Mode, which uses more computational resources to handle complex tasks.
According to Musk, Grok 3 is the smartest AI on earth, so smart that it is a little scary. It is designed to redefine the way problems are solved using AI. The model has more than 10 times the computational power of its predecessor, making it capable of handling more demanding tasks. It aims to serve as a versatile tool for various industries, enabling
everything from scientific research to technical problem-solving. xAI is also launching a Grok AI agent product called Deep Search, which, according to the company, is a next generation search engine.
The name “Grok” is inspired by the character Grok in Robert Heinlein’s science fiction novel Stranger in a Strange Land, who is raised in Mars and is meant to convey deep understanding and empathy.
Challenging others in the fray
Grok 3 is designed to challenge dominant players like OpenAI in the AI landscape. Musk’s ambition for xAI has always been to produce AI that is more efficient, more transparent, and smarter than current systems. While OpenAI’s GPT-4 and DeepSeek’s V3 are recognised for their capabilities, Grok 3 promises to offer distinct features like enhanced reasoning and powerful computational modes that can solve complex issues. By introducing Big Brain Mode, Grok 3 sets itself apart by dedicating additional resources to tackle tough challenges.
This makes it particularly appealing for industries requiring detailed analysis and creative solutions. Musk has emphasized that the goal of Grok 3 is to build AI that can think and reason in ways more aligned with human intelligence. The competition will only grow as independent benchmarks are carried out, but Grok 3 has the potential to establish itself as a strong player in this field, experts said.
How can this be accessed by users?
The latest variant of Grok can be accessed through Musk’s social media platform, X (formerly known as Twitter). Users who wish to use the model’s advanced capabilities can subscribe to the X Premium+ service or SuperGrok tier. SuperGrok, which costs $30 per month, grants access to Grok 3’s advanced reasoning modes and computational power, offering enhanced performance for more demanding tasks.
Post the launch of Grok 3, X also almost doubled the pricing of X Premium +. In India, it is available for Rs 3,470 per month vs Rs 1,750 per month earlier. Musk has stated that this subscription model ensures greater accessibility to AI tools and that Grok 3 is designed for those who need more powerful and efficient problem-solving capabilities. Grok 2 can be used by users currently without any fee. Musk is also planning to make Grok 2 open source. The strategy of open sourcing only the previous version of the model, rather than the current one, helps xAI protect its value proposition.
Cost and GPU training details
Grok 3 was trained using 200,000 Nvidia H100 GPUs.The model was developed in two stages. Initially, 122 days of training was done on 100,000 GPUs, followed by 92 days of scaling up to 200,000 GPUs, xAI said. With 200,000 GPUs used, the total cost for the GPUs alone could range from $6 billion to $8 billion, assuming all were purchased outright, according to sector experts. In comparison, DeepSeek’s reported cost of building the model was $6 million, which many industry executives cite as misleading.
Notably, Grok 2 was trained on 100,000 GPUs. Grok 3’s enormous computational power required for training is reflected in the model’s capabilities, allowing it to process vast amounts of data and perform complex tasks with greater speed and accuracy. Musk has noted that its computing power is orders of magnitude more efficient compared to previous models. The three key enhancements involve synthetic dataset capabilities, self-correction functions, and reinforcement learning mechanisms.
Reaction from early users
Even as xAI claims that Grok 3 outperforms several competitors in benchmarks, some users argue that these benchmarks might not fully reflect practical usability. Posts on X show excitement for the new possibilities Grok 3 brings, but also a wait-and-see approach, with some users expressing that they prefer other AI models for certain tasks due to Grok 3’s lack of deep understanding in problem-solving. There is a skepticism regarding its performance consistency and real-world applicability. According to some, Grok’s responses might sometimes be too lengthy, or lack in clarity compared to other models like OpenAI’s GPT-4o. Many users, however, have also appreciated its ability to generate code, analyse cultural references, and provide insightful responses across a range of subjects.
