xAI, Elon Musk's artificial intelligence company, has launched Grok 4, the latest iteration of its AI chatbot, marking a significant leap in AI technology. The new model is accessible through a premium "Pro" subscription tier on X (formerly Twitter), named SuperGrok Heavy, priced at $300 per month. This subscription unlocks advanced AI capabilities, targeting power users and developers who demand more than standard chatbot tools can offer.
Grok 4 is designed to compete with top-tier AI models like OpenAI's ChatGPT and Google's Gemini. xAI claims that Grok 4 demonstrates "frontier-level performance" across several benchmark evaluations and has "PhD-level intelligence". Musk has even suggested Grok 4 has the potential to move beyond human limitations in knowledge discovery, potentially uncovering new technologies by the end of 2025 and even "new physics" within the next two years.
Grok 4 boasts enhanced reasoning, scientific problem-solving, and real-time information access. Trained on over 200,000 GPUs using xAI's Colossus supercomputer, it achieves ten times the reasoning power of Grok 3. The model utilizes reinforcement learning to tackle complex and novel problems across multiple subjects.
Two versions of Grok 4 are available: a generalist model for tasks like writing, translating, and research, and a developer model tailored for coding assistance, debugging, and integration with development environments. The "Grok 4 Code" variant offers intelligent code completion, debugging assistance, and optimization suggestions, making it a powerful coding companion with real-time IDE integration. Musk stated Grok 4 can debug entire codebases, claiming it works better than Cursor, an AI-powered code editor.
Grok 4 supports multiple input types such as text and images, with vision and image-generation features expected soon. Grok 4 has a 256k token context window accessible via API. It also features enhanced voice capabilities, delivering faster and more natural responses.
Grok 4 Heavy is a multi-agent version where multiple AI agents run in parallel, comparing notes to deliver improved accuracy. This "study group" approach allows the model to solve more problems, especially on complex benchmarks. The SuperGrok Heavy subscription plan grants users early access to Grok 4 Heavy and upcoming product features, including an AI coding model, a multi-modal agent, and a video generation system.
Grok 4 has demonstrated remarkable performance in benchmarks. On "Humanity's Last Exam," a difficult AI benchmark, Grok 4 Heavy solved over 50% of the text-only subset of problems. On the ARC-AGI V2, Grok 4 obtained 15.9%, doubling the previous score. xAI claims that Grok 4 outperforms competitors like OpenAI's o3 and Google's Gemini 2.5 Pro on certain benchmarks.
Grok 4 is accessible via xAI's API, featuring a 256k token context window, real-time search capabilities, multimodal support, and enterprise-grade security. Grok 4 will be deployed across hyperscalers and enterprise environments, with ongoing updates planned.
Grok 4's pricing structure includes input tokens at $3.00 per 1M tokens, cached input tokens at $0.75 per 1M tokens, and output tokens at $15.00 per million tokens. The subscription model offers three tiers: Basic (free with limited Grok 3 access), SuperGrok ($300 annually with Grok 4 access), and SuperGrok Heavy ($3,000 annually with Grok 4 Heavy preview access).
The launch of Grok 4 followed controversy over offensive posts generated by Grok 3. xAI addressed these concerns, stating they are actively working to remove inappropriate posts and ban hate speech. Musk attributed the remarks to shortcomings in the AI's ability to filter human input and stated that this is being addressed.
