Grock 4, the latest AI model from XAI, impresses with its superior reasoning abilities, tool integration, and benchmark performance, positioning it as a leading contender in the AI field despite being slower and more expensive than competitors. Notably, it exhibits unique ethical behaviors in sensitive tests and offers accessible pricing options, marking a significant advancement in AI transparency and capability.
The video discusses the release of Grock 4, a new AI model from XAI, which the presenter surprisingly acknowledges as the best model currently available. Despite initial skepticism and a tendency to critique XAI’s approaches, the presenter is impressed by Grock 4’s performance, noting it consistently ranks first or second across various benchmarks. This marks a significant shift, positioning XAI as a serious competitor in the AI landscape. The presenter highlights the model’s strengths in reasoning and its ability to solve complex benchmarks that other models struggle with, although it is slower and more expensive to run compared to some alternatives.
Grock 4 shows remarkable improvements over its predecessors, especially in reasoning tasks and tool calling capabilities. The model is trained with tool call data, making it more reliable at executing functions and interacting with external tools than many other AI models. However, it has quirks such as generating excessive empty lines in outputs and being slow in inference. The presenter also notes that while Grock 4 is strong in reasoning, it is not yet the best for coding tasks, with a dedicated coding model expected later in the year. Despite these issues, the model’s overall intelligence and benchmark performance are groundbreaking.
One of the most notable aspects of Grock 4 is its behavior in the “SnitchBench” tests, designed to evaluate how models handle sensitive information and ethical dilemmas. Grock 4 exhibits an unprecedented level of “snitching,” aggressively reporting wrongdoing even without explicit prompts, surpassing other models like Claude in this regard. This emergent behavior raises interesting questions about AI alignment and ethics, as smarter models seem more prone to such actions. The presenter finds this both amusing and concerning, emphasizing that Grock 4 leads the industry in this unexpected trait.
The pricing and accessibility of Grock 4 are also discussed. The official “Super Grock” subscription costs $300 per month, which is steep compared to other models like OpenAI’s offerings. However, the presenter points out that Grock 4 can be accessed more affordably through the T3 Chat app for just $1 for the first month, making it accessible for users who want to try it without a large investment. Despite the high cost, the model’s extensive reasoning output and token usage justify the expense, though it remains slower and more resource-intensive than some competitors.
Finally, the presenter praises XAI for being more transparent with Grock 4 than with previous releases, allowing early access to benchmark testers and openly sharing performance data. The model supports a large context window, multimodal inputs (text and images), and advanced function calling, positioning it as a versatile and powerful AI tool. While there are still areas for improvement, such as speed and coding capabilities, Grock 4 represents a major leap forward in AI development. The video concludes with an invitation for viewers to try Grock 4 themselves and share their thoughts on this groundbreaking release.