xAI Launches Grok 4

On Wednesday evening, Elon Musk's artificial intelligence company xAI officially released its latest flagship AI model, Grok 4, and simultaneously launched a high-end subscription service SuperGrok Heavy, with a monthly fee of up to $300, making it the most expensive among the current mainstream AI service providers..

As a response to competitors such as OpenAI's ChatGPT and Google Gemini, Grok has continued to deeply integrate with the social platform X (formerly Twitter) in recent years, and further strengthened its connection with xAI's official acquisition of X. This has also made Grok's performance, especially in content generation and public opinion response, attract much attention from users and the industry.

Musk said: Grok 4 is already above PhD level, "lack of common sense is only temporary".

In the live broadcast that night, Musk appeared in a leather jacket and introduced the new capabilities of Grok 4 alongside the xAI team. He stated that Grok 4's academic abilities in various disciplines "fully surpass the doctoral level." Although it still has room for improvement in common sense reasoning and creative discovery, he said, "it's only a matter of time."

In this release, xAI launched two versions of the model simultaneously:

• Grok 4: the standard flagship model, supporting image analysis and question answering.
• Grok 4 Heavy: built on a "multi-agent system" where multiple AIs work on a problem simultaneously, compare their outputs, and produce the optimal solution—an approach inspired by the idea of an “AI learning group.”

According to the company, this collaborative architecture is the core reason Grok 4 Heavy performs well in complex tasks.

However, behind the technological progress, xAI has also encountered considerable turmoil.

On the same day as the press conference, Linda Iaccarino, CEO of X Lab, announced her resignation, ending her nearly two-year term.

Just a few days prior, Grok’s automated account made headlines when it accidentally posted a comment in support of Hitler while responding to a user’s anti-Semitic statement. The incident caused public uproar. xAI swiftly restricted the account and deleted the offensive content.

In response, xAI quietly removed a controversial system prompt that had previously instructed the AI not to avoid making "politically incorrect" statements.

Despite these controversies, Musk and the executive team chose not to address them during the press event, instead focusing on highlighting the performance and capabilities of Grok 4.

xAI stated that Grok 4 achieved strong results on several authoritative benchmarks, including the well-known “Humanity’s Last Exam”—a highly challenging test designed to assess AI performance across thousands of real-world problems in math, science, and the humanities.

• Grok 4 (without tools) score: 25.4%

         Higher than Gemini 2.5 Pro (21.6%) and OpenAI o3 (21%)

• Grok 4 Heavy (using tools) score: 44.4%

         Also surpassing Gemini 2.5 Pro (26.9%) using tools

In another highly professional test ARC-AGI-2, Grok also scored 16.2%, nearly twice that of Claude Opus 4, setting a new record for the test.

To link up with the launch of the Grok 4 Heavy model, xAI simultaneously released a new high-end subscription service: SuperGrok Heavy. This plan is priced at $300 per month, which is currently the most expensive personal subscription plan among mainstream AI vendors. Compared with the high-end services of platforms such as OpenAI, Anthropic and Google, SuperGrok Heavy is obviously more aggressive in price and more "experimental".

This subscription service is aimed at professional user groups who have extremely high requirements for cutting-edge performance. Subscribers to SuperGrok Heavy will receive a series of priority benefits, including the right to use the Grok 4 Heavy model first, and be the first to experience the latest features that will be launched soon. This means that before the official release, some features will be tested and iterated on a small scale in this subscription tier, and users will also have the opportunity to participate in feedback and interaction at the first time.

xAI also disclosed the rhythm of the next product update, outlining a clear future roadmap for subscribers. According to the plan, the AI coding model will be launched in August to provide developers and engineers with stronger code generation and collaboration capabilities; in September, a multimodal intelligent system will be launched to integrate text, image, voice and other inputs to create an AI interactive experience that is closer to human cognition; in October, xAI will officially release a model that supports video generation, completing a key link in the field of content generation.

In addition to personal subscriptions, xAI is also making full efforts to lay out enterprise services. Grok 4 has been opened to developers through APIs to support the construction of various application scenarios; although its enterprise department has only been established for two months, it has planned to cooperate with large institutions through cloud platforms to promote the enterprise implementation of Grok.

Although Grok 4 has made breakthroughs in multiple AI benchmarks and its model capabilities have attracted the attention of many developers, the controversial brand image issue remains unresolved.

For enterprise customers, technical strength is critical, but stability, security, and matching of values are equally important. If xAI wants to truly become a strong competitor to ChatGPT, Claude, and Gemini, it needs to move more steadily and respond more forcefully.