xAI has released a preview of its new Grok-2 artificial intelligence model, which is significantly superior to the previous Grok-1.5. The new model demonstrates advanced abilities in communication, coding and reasoning. In addition to Grok-2, a more compact version – Grok-2 mini – was presented.

Internal tests show that the Grok-2 scores higher in specialized benchmarks than the Claude 3.5 Sonnet and the GPT-4-Turbo. The Grok-2 model excels in areas such as academics, math, coding, and tasks that require visual perception and information analysis. The Grok-2 and Grok-2 mini are currently in beta testing on the X platform and are available to users with a paid Premium subscription.

Developers plan to provide access to both models via an enterprise API later this month. The Grok-2 model is versatile and supports image generation. The first users have already noted the lack of censorship, including the ability to create images with images of political figures. According to previously published information, the Grok-2 model was trained on 24,000 Nvidia H100 accelerators.

Elon Musk intends to increase investments in computing centers in order to bring Grok to a leading position in the market. To train the next version of AI, it is planned to launch the world’s largest computing cluster using 100,000 Nvidia H100 graphics processors.

Previously, the function of creating images using the Grok generator became available. users of the X Premium+ subscription, which costs $16 per month (or $22 when signing up through the app). It is important to note that this subscription level allows users to enjoy ad-free content.

Grok, while still in its early stages of development, is showing progress in imaging. Although they look quite realistic, there are some features that indicate their artificial origin. These include, for example, airbrushed details and other typical AI errors, such as the appearance of random, unrecognizable shapes in images.

Despite these shortcomings, the launch of the Grok-2 and Grok-2 Mini shows that the company is serious about continuing to improve its models. This could impact the generative artificial intelligence market, creating new opportunities and competition in the field.