xAI Debuts Grok 4 and Grok 4 Heavy, Signals Ambitious Roadmap Through October

xAI released Grok 4 and Grok 4 Heavy this week, introducing two new variants of its proprietary large language model. Built on what the company calls “Foundation Model level 6,” the models show marked performance gains, including a 25.4 % score for Grok 4 on Humanity’s Last Exam and a 44.4 % score for Grok 4 Heavy when tools are enabled. These figures place Grok ahead of Google’s Gemini 2.5 Pro and OpenAI’s o3, at least in raw benchmark terms.

Elon Musk described the new model as operating at “PhD-level” knowledge across disciplines. Despite the fanfare, Grok 4 Heavy is only available to early testers through a newly introduced $300/month “SuperGrok Heavy” subscription plan. The premium tier is still in its early access phase, targeting power users and developers interested in faster access and model experimentation.

The company also revealed that training has already begun on the next generation architecture: Foundation Model level 7. xAI expects training to complete in a matter of weeks, suggesting that this release is a staging ground ahead of a more powerful update. If level 6 laid the groundwork, level 7 may define the competitive ceiling for Grok in 2025.

In a forward-looking roadmap shared during the launch, xAI outlined a timeline for rapid expansion of Grok’s capabilities. A specialized coding model is scheduled for August. A multimodal agent—likely capable of combining text, image, and possibly audio inputs—is planned for September. In October, the company aims to release a video generation model, marking a major leap toward media synthesis and creative tooling.

This signals that xAI is not just trying to catch up to rivals. It is attempting to leapfrog them on roadmap speed and feature deployment. Unlike OpenAI’s relatively staged releases or Google’s tightly integrated Gemini stack, xAI is moving in aggressively iterative bursts with a clear vertical in mind: fast, direct, frontier-grade AI for people who want full access.

For now, Grok remains embedded inside X (formerly Twitter), but developer access via API ensures the technology can be deployed more broadly. The heavy emphasis on model transparency, fast shipping, and internet-native tone suggests xAI is building not just an assistant, but an ecosystem aligned with its platform’s culture.

Whether Grok can scale beyond early adopters will depend on the success of its next-gen foundation model. The current $300/month “SuperGrok Heavy” plan signals xAI’s belief that there is a market for premium, early-access AI tools. But for now, Musk’s AI venture is signaling that it’s not here to play catch-up. It’s here to build fast, launch loud, and push the market forward at internet speed.

Ah yes, nothing says “democratizing AI” like a $300/month price tag. Elon Musk’s xAI has launched Grok 4 and Grok 4 Heavy with all the subtlety of a Tesla Cybertruck driving through a library. The models claim PhD-level intelligence, but the real thesis here is about who gets early access to frontier AI: the meme-literate elite willing to pay for the privilege.

The roadmap is ambitious — coding next month, multimodal by September, and video generation in October. If they hit those targets, xAI won’t just be building a chatbot. They'll be building a full-stack Muskverse with its own language, media, and logic. Of course, that also assumes they finish training Foundation Model level 7 without crashing into another hype wall.

For now, Grok isn’t replacing ChatGPT or Gemini for the masses. But it’s not trying to. It’s courting the impatient, the wealthy, and the always-on crowd who don’t just want AI — they want their AI, fast, loud, and unapologetically weird.