Elon Musk and the Evolution of AI: From OpenAI to xAI's Grok 3
Elon Musk, a prominent figure in technology and entrepreneurship, initially invested in OpenAI at its inception in 2015. However, ideological differences and strategic shifts at OpenAI led Musk to sever his connections with the company, which he felt had deviated from its original non-profit aims. In pursuit of his vision for artificial intelligence, Musk founded his own AI venture, xAI, culminating in the development of a large language model (LLM) named Grok. Recently, xAI launched a new iteration, Grok 3, making significant waves in the competitive chatbot landscape.
Unveiling Grok 3
On a significant Monday launch event, Elon Musk introduced xAI's latest breakthrough, the Grok 3 family of AI models, through a live stream. Grok 3 boasts a tenfold increase in training compared to its predecessor, Grok 2, powered by xAI's state-of-the-art data center located in Memphis, Tennessee, housing an impressive array of 200,000 GPUs.
During the live stream, Musk expressed his excitement about Grok 3's capabilities, describing it as not just an incremental improvement, but an exponential leap beyond Grok 2. The unveiling also included a unique reasoning model, building upon Grok 3's foundation, designed to enhance cognitive abilities by processing information more thoroughly, akin to existing reasoning models from competitors like OpenAI’s o1 and o3.
Competing in the AI Arena
The Grok 3 models are positioned to rival leading AI models, with the standard Grok 3 directly competing against OpenAI's GPT-4o and Google's Gemini, while Grok 3 Reasoning seeks to challenge models such as o3-mini (high), o1, and Deepseek-R1. Within just 24 hours of its release, xAI's Grok 3 started dominating performance benchmarks and leaderboards, indicating a strong competitive edge.
Performance and Capabilities
The pre-training phase for Grok 3 concluded in January, and despite ongoing training, the model has already surpassed competitors in several AI benchmarks. Tests such as AIME 24 (mathematical reasoning), GPQA (scientific proficiency), and LCB Oct-Feb (coding capabilities) have highlighted Grok 3's superior performance.
Grok 3's reasoning models, although still in beta, have demonstrated impressive competitiveness with top-tier models in the market, as evidenced during the live stream's shared results. These models performed admirably across extensive benchmarks, showcasing their potential to disrupt existing AI hierarchies.
Dominance in Chatbot Arena
Screenshot by Sabrina Ortiz/ZDNET
Beyond technical benchmarks, Grok 3 quickly ascended the ranks in the Chatbot Arena, a platform where users evaluate language models by direct comparison. An early version of Grok 3, codenamed "chocolate," achieved a historic milestone, becoming the first model to exceed a 1400 score and claiming the top spot across all categories.
xAI's DeepSearch Initiative
To address the growing demand for AI with agentic capabilities, xAI introduced DeepSearch, a feature paralleling OpenAI and Google's deep research functionalities. DeepSearch enables users to pose questions and receive comprehensive answers, simulating a thorough cognitive process. This feature allows users to observe the AI's thought process, enhancing transparency and user control over the generated results.
| Feature | Grok 3 | OpenAI's GPT-4o | Google's Gemini |
|---|---|---|---|
| Training Volume | 10x more than Grok 2 | Established benchmarks | Established benchmarks |
| Center Technology | Memphis-based, 200,000 GPUs | Advanced infrastructure | Advanced infrastructure |
| Reasoning Model | Enhanced cognitive processing | o1 and o3 models | Deep processing capabilities |
Access and Subscription
Starting today, users can access select Grok models in beta form. Grok 3 is available through the X Premium+ subscription, which offers access to new features, increased usage limits, DeepSearch functionality, and advanced reasoning modes. This subscription is priced at $40 per month, an increase from its previous rate of $22.
xAI also launched a new subscription tier, SuperGrok, targeted at enthusiasts seeking early access to cutting-edge capabilities. While the pricing for this tier remains undisclosed, it is anticipated to rival competitive services, such as OpenAI's Pro subscription, priced at $200 per month.
Future Developments and User Engagement
Elon Musk advises users seeking the most refined version of Grok 3 to wait an additional week, by which time a new voice integration feature is expected to be available. For users interested in participating in the Chatbot Arena and testing Grok 3, they can visit the platform, select "early-grok-3" from the options, and initiate interactions with the AI model.
