xAI hosted a livestreamed event on the X platform to launch Grok 4, the latest version of its artificial intelligence AI language model.
Many viewers had enthusiastic reactions. "Grok 4 is basically AGI," Beff Jezos posted to X.
The launch event highlighted Grok 4’s significant advancements over Grok 3, positioning it as a leading competitor to models like OpenAI’s GPT-4o and Anthropic’s Claude. The livestream featured demonstrations and insights into Grok 4’s capabilities, drawing attention from developers, tech enthusiasts, and industry observers.
The event showcased Grok 4’s two variants: a general-purpose model and Grok 4 Code. The general model excels in enhanced reasoning, handling complex tasks like mathematical and scientific queries with high accuracy. It supports a 256k token context window, allowing for extended conversations, and offers structured output for precise responses.
Described as "the world's most powerful AI model" and "the same model physicists use," Grok 4 emphasizes advanced logical reasoning, text generation, and multimodal capabilities (handling text, images, and vision, with video understanding planned). Elon Musk claimed it's "better than almost all PhDs in all fields" on academic tests, though he noted limitations in common sense and real-world invention.
Grok 4 Code, tailored for developers, integrates with tools like Cursor for code generation, debugging, and contextual programming support, making it a powerful coding companion.
Demonstrations and Features
Live demos illustrated Grok 4’s real-time data search via X integration, enabling up-to-date responses, and its enterprise-grade security through the xAI API, appealing to professional users. Benchmarks presented during the stream showed Grok 4 leading with an Artificial Analysis Intelligence Index of 73, surpassing OpenAI’s o3 and Google’s Gemini 2.5 Pro. It achieved a 24% score on Humanity’s Last Exam and an 88% on GPQA Diamond, setting new records for academic and reasoning tasks. The event emphasized Grok 4’s speed (75 output tokens/second) and its training on xAI’s Colossus supercomputer, a 200,000-NVIDIA-GPU cluster, which boosted its performance.
A key highlight was the introduction of Grok 4 Heavy, a specialized model designed for highly complex, compute-intensive tasks like advanced scientific simulations and large-scale data analysis. "Grok 4 Heavy is already ASI level," enthused Beff Jezos.
A new $300-per-month "SuperGrok Heavy" plan was unveiled for early access to Grok 4 Heavy and upcoming features.
Mainstream news coverage of Grok 4 is beginning to appear. Here's a TechCrunch article. This is a developing story, to be continued.