Grok 4 Called Equivalent to PhD, Offers $300 Subscription

Grok 4 was developed as the successor to Grok 3, skipping the planned “3.5” version release. This model comes in two main variants: regular Grok 4 and Grok 4 Heavy. The “Heavy” variant is a multi-agent version promised to deliver higher performance. One of Grok 4's standout features is multimodal capability: it can analyze images while processing text, similar to competitors like ChatGPT and Gemini.

In their presentation, xAI claimed that Grok 4 demonstrates state-of-the-art performance on various AI benchmarks. For example, in the Humanity’s Last Exam test – a complex exam measuring AI capabilities in mathematics, humanities, and science – Grok 4 without the aid of “tools” scored 25.4%, outperforming Google Gemini 2.5 Pro (21.6%) and OpenAI model o3 (21%).

Even with additional “tools,” Grok 4 Heavy achieved a score of 44.4%, much higher compared to Gemini with tools (26.9%). On the visual ARC-AGI-2 test assessing visual pattern solving capabilities, Grok 4 also excelled: a score of 16.2% made it the new champion among commercial models, nearly doubling the next model (Claude Opus 4).

This improvement was plainly summarized by Elon Musk at the launch: “In academic questions, Grok 4 is better than PhD level in every subject, no exceptions,” he said.

Musk also acknowledged that weaknesses remain, for instance, sometimes Grok lacks common sense or has not yet created new technology independently – but he is confident it's only a matter of time. Besides intelligence improvements, xAI's product roadmap includes a specialized coding model ready for launch in August, a multi-modal agent in September, and a generative video model in October, showing Musk’s vision to continue expanding Grok’s capabilities in various areas.

Alongside Grok 4, xAI introduced a new subscription package called SuperGrok Heavy costing $300 per month. This is the most expensive package ever offered by xAI, even surpassing the ultra-tier fees of other major AI providers (OpenAI, Google, Anthropic). SuperGrok Heavy subscribers get early access (preview) to Grok 4 Heavy and new experimental features before the general public, including the potential for early access to upcoming xAI AI products. For example, the company mentioned that subscribers of this package could have first access to coding models, advanced agents, or upcoming video models.

The live launch event for Grok 4 on July 9 featured Elon Musk and xAI leadership focusing their presentation on the new model’s performance and capabilities. Musk illustrated how far Grok 4 has advanced compared to previous versions, including the success of Grok 4 Heavy in solving advanced benchmarks. He emphasized that the xAI team has worked hard to address previous errors; earlier Grok had been criticized for politically biased answers, which Musk once called a “major fail”, but he assured that his team has fixed this in the latest version.