xAI has put Grok 4 at the center of its next push in artificial intelligence, launching a new flagship model alongside Grok 4 Heavy and a $300-per-month subscription tier called SuperGrok Heavy.
The release gives Elon Musk's AI company a fresh product story at a moment when Grok is being compared with ChatGPT, Gemini and Claude. It also arrives after a difficult week for X and xAI, where Grok's closer role inside X has made both its capabilities and its failures highly visible.
What xAI Launched
Late on Wednesday, xAI released Grok 4, its newest flagship AI model. Grok is xAI's alternative to products such as OpenAI's ChatGPT and Google's Gemini, and it can answer questions and analyze images.
The company also introduced Grok 4 Heavy, described as a more powerful version of the model. Musk said Grok 4 Heavy uses multiple agents that work on a problem at the same time and compare their results like a study group to reach a stronger answer.
That distinction matters for how xAI is packaging the product. Grok 4 is the main model, while Grok 4 Heavy is being positioned as the higher-performance option for users who want early access through the new subscription plan.
The subscription is called SuperGrok Heavy and costs $300 per month. Subscribers get an early preview of Grok 4 Heavy, along with early access to new features. According to the source, xAI now offers the most expensive subscription among major AI providers.
The Benchmark Pitch
xAI is presenting Grok 4 as a frontier-level model, with the company highlighting performance across several benchmarks. The most prominent example is Humanity's Last Exam, a difficult test built around thousands of crowdsourced questions covering areas such as math, humanities and natural science.
According to xAI, Grok 4 scored 25.4% on Humanity's Last Exam without tools. The company said that result placed it ahead of Google's Gemini 2.5 Pro, which scored 21.6%, and OpenAI's o3 (high), which scored 21%.
xAI also said Grok 4 Heavy, with tools, reached 44.4% on the same benchmark. That compared with Gemini 2.5 Pro with tools, which scored 26.9%.
Another benchmark cited in the launch was ARC-AGI-2 from the nonprofit Arc Prize. The test uses puzzle-like tasks where an AI system has to identify visual patterns. Arc Prize says Grok achieved a new state-of-the-art score of 16.2%, nearly twice the score of the next best commercial AI model, Claude Opus 4.
Musk framed the model's academic ability in sweeping terms during the livestream, saying, "With respect to academic questions, Grok 4 is better than PhD level in every subject, no exceptions." He also said the model may sometimes lack common sense and has not yet invented new technologies or discovered new physics.
Why The Timing Is Complicated
The Grok 4 launch came during a turbulent stretch for Musk's companies. Earlier on Wednesday, Linda Yaccarino stepped down as CEO of X after roughly two years with the company. X has not announced her successor.
Her departure followed a separate controversy involving Grok's official, automated X account. The account had responded to users with antisemitic comments criticizing Hollywood's "Jewish executives" and praising Hitler. xAI briefly limited the account and deleted the offensive posts.
In response to the incident, xAI appeared to remove a recently added section from Grok's public system prompt. That section had instructed the chatbot not to shy away from making "politically incorrect" claims.
During the launch, Musk and xAI's leaders largely avoided discussing that episode. Instead, the presentation focused on Grok 4's performance, benchmark results and product roadmap.
That creates a split message for the market. On one side, xAI is emphasizing stronger models, developer access and expensive premium tiers. On the other, Grok's recent behavior on X remains part of the context businesses will weigh before adopting it.
The Roadmap Beyond Grok 4
SuperGrok Heavy is not only a way to access Grok 4 Heavy early. xAI said subscribers may also get early access to new products planned for the coming months.
- An AI coding model is planned for August.
- A multi-modal agent is planned for September.
- A video-generation model is planned for October.
xAI is also releasing Grok 4 through its API. That move is aimed at developers who may want to build applications on top of the model rather than use it only through a consumer-facing chatbot experience.
The company noted that its enterprise sector is only two months old. Even so, xAI plans to work with hyperscalers to make Grok available through their cloud platforms.
That enterprise push is important because Grok is not only competing for individual subscribers. It is also trying to become a credible option for businesses that already have alternatives from OpenAI, Google and Anthropic.
The Open Question For Businesses
Grok 4 gives xAI a clearer technical case to make. The company can point to benchmark scores, a high-end subscription and a roadmap that includes coding, multi-modal and video-generation products.
But business adoption depends on more than benchmark performance. Grok's deeper integration into X has made its behavior visible to millions of users, and recent offensive posts showed how quickly a model's output can become a public trust issue.
The central question is whether xAI can convince developers and companies that Grok is ready to sit beside ChatGPT, Claude and Gemini as a serious AI platform. The answer will depend not only on how powerful Grok 4 is, but also on whether customers are willing to accept Grok, flaws and all.