xAI, Elon Musk’s artificial intelligence company, released its latest flagship AI model, Grok 4, and introduced a new $300-per-month AI subscription plan, SuperGrok Heavy, on Wednesday.
Grok serves as xAI’s direct competitor to models such as OpenAI’s ChatGPT and Google’s Gemini, possessing the capability to analyze images and respond to questions. In recent months, Grok has become more deeply integrated into X, the social network owned by Elon Musk, which xAI recently acquired. This integration has, however, brought Grok’s occasional misbehavior to the attention of a wide user base.
Introducing Grok 4, the world’s most powerful AI model. Watch the livestream now: https://t.co/59iDX5s2ck
— xAI (@xai) July 10, 2025
Expectations for Grok 4 are substantial, as xAI’s newest AI model will be directly compared against OpenAI’s forthcoming AI model, GPT-5, which is anticipated to launch later in the summer. During a livestream held on Wednesday night, Elon Musk stated, “With respect to academic questions, Grok 4 is better than PhD level in every subject, no exceptions. At times, it may lack common sense, and it has not yet invented new technologies or discovered new physics, but that is just a matter of time.”
The launch of Grok 4 occurred amidst a challenging week for Elon Musk’s various companies. Earlier on Wednesday, Linda Yaccarino resigned from her position as CEO of X after approximately two years with the company. X has not yet announced her successor. Yaccarino’s departure followed an incident where Grok’s official, automated X account responded to users with antisemitic comments, criticizing Hollywood’s “Jewish executives” and praising Hitler. xAI temporarily restricted Grok’s account and removed the offensive posts. Following this incident, xAI appeared to have removed a recently added section from Grok’s public system prompt, which is a list of instructions for the AI chatbot to follow, that had previously instructed it not to shy away from making “politically incorrect” claims. Musk and xAI’s leadership largely avoided discussing this specific incident, instead concentrating on Grok 4’s performance and capabilities during their public statements.
Türkiye blocks Grok nationwide over alleged insults
xAI launched two distinct models on Wednesday: Grok 4 and Grok 4 Heavy. Grok 4 Heavy is described as the company’s “multi-agent version,” engineered to provide increased performance. Musk explained that Grok 4 Heavy operates by spawning multiple agents to collaboratively work on a single problem. These agents then compare their respective findings, functioning “like a study group,” to determine the most accurate or optimal answer.
xAI asserts that Grok 4 demonstrates frontier-level performance across several benchmarks. One such benchmark is Humanity’s Last Exam, a demanding test designed to assess an AI’s capacity to answer thousands of crowdsourced questions spanning subjects such as mathematics, humanities, and natural sciences. According to xAI, Grok 4 achieved a score of 25.4% on Humanity’s Last Exam without the use of “tools,” thereby outperforming Google’s Gemini 2.5 Pro, which scored 21.6%, and OpenAI’s o3 (high), which scored 21%. Furthermore, xAI claims that Grok 4 Heavy, when utilizing “tools,” was able to attain a score of 44.4%, surpassing Gemini 2.5 Pro with tools, which achieved 26.9%.
The nonprofit Arc Prize has confirmed that Grok has achieved a new state-of-the-art score on its ARC-AGI-2 test. This benchmark is another difficult assessment consisting of puzzle-like problems where an AI must identify visual patterns. Grok scored 16.2% on this test, which is nearly double the score of the next best commercial AI model, Claude Opus 4.
In conjunction with the release of Grok 4 and Grok 4 Heavy, xAI introduced its most expensive AI subscription plan to date, a $300-per-month offering named SuperGrok Heavy. Subscribers to this plan will receive an early preview of Grok 4 Heavy, as well as early access to forthcoming new features. This plan is comparable to the ultra-premium tiers offered by other major AI providers such as OpenAI, Google, and Anthropic. However, xAI now provides the most expensive subscription among these leading AI developers. SuperGrok Heavy subscribers may also gain early access to specific new products that xAI intends to launch in the coming months. The company stated on Wednesday that an AI coding model is scheduled for release in August, a multi-modal agent in September, and a video generation model in October.
xAI is making Grok 4 available through its API, with the intention of encouraging developers to build applications utilizing the model. The company noted that xAI’s enterprise sector is only two months old. Nevertheless, it plans to collaborate with hyperscalers to ensure Grok is accessible through their respective cloud platforms. Despite Grok’s demonstrated frontier-level performance on benchmarks, xAI may face challenges in overcoming public perception issues stemming from its recent mishaps as it endeavors to position Grok to businesses as a viable competitor to ChatGPT, Claude, and Gemini.
