XAI, Elon Musk’s artificial intelligence company, has launched its latest AI model Grok 3, marking a significant development in the competitive AI landscape. The new model family was unveiled during a livestreamed presentation, with Musk claiming it represents a major advancement over its predecessor, Grok 2.
Technical Infrastructure and Capabilities
The model’s development relied on substantial computing power, utilizing approximately 200,000 GPUs in a Memphis data center. According to xAI, Grok 3 required ten times more computing power for training compared to Grok 2. The company claims the model outperforms competitors like OpenAI’s GPT-4o and Google’s Gemini on several benchmarks, including mathematics (AIME) and scientific reasoning (GPQA).
The Grok 3 family includes multiple variants:
- Grok 3 (main model)
- Grok 3 mini (faster but less accurate)
- Grok 3 Reasoning
- Grok 3 mini Reasoning
A notable feature is the new DeepSearch capability, which combines web searching with reasoning abilities to analyze information from multiple sources. The system includes specialized modes for problem-solving, including a “Think” function and a “Big Brain” mode for complex queries.
Availability and Pricing
Initial access to Grok 3 will be available to X Premium+ subscribers ($50 monthly). Additional features will be accessible through a new SuperGrok subscription, reportedly priced at $30 monthly or $300 annually. Enterprise API access is planned for release in the coming weeks.
Independent experts, including former Tesla Director of AI Andrej Karpathy, have confirmed some of xAI’s performance claims while noting certain limitations, such as occasional issues with citations and ethical reasoning tasks.
The company plans to open-source its previous model, Grok 2, once Grok 3 reaches stability, maintaining its practice of releasing older versions as newer ones mature.
Sources: TechCrunch, Tom’s Hardware, VentureBeat