- Official launch of Grok 4, the IA from xAI, with improvements in reasoning and speed
- It competes directly against GPT-4.5, Gemini 1.5 and Claude 4, surpassing key benchmarks
- Two versions available: Grok 4 standard and Grok 4 Heavy, the latter with a premium subscription
- Recent controversies have prompted changes to the model for filtering offensive responses.
La Artificial Intelligence has returned to the center of attention after the Grok 4 launch announcement, the latest evolution of the chatbot developed by xAI, the company powered by Elon MuskIn less than half a year after the arrival of Grok 3, the company has taken a step forward with what promises to be one of the AI most powerful, fastest and most advanced in the industry, in a context where competition with OpenAI, Google and Anthropic is getting more and more ferocious.
With a live presentation followed by millions of people and Musk's statements full of expectations, the technological ecosystem is wondering what this progress really means for the generative AI market, both on a technical level and in the way we interact with these tools.
New capabilities and two versions: Grok 4 and Grok 4 Heavy
The leap forward with Grok 4 is not limited solely to response speed or text comprehension. xAI has opted to equip its new model with advanced capabilities for real-time reasoning and multimodal analysis. (it's capable of understanding both text and images) and an architecture optimized for complex tasks. All of this is possible thanks to Colossus, xAI's supercomputer located in Memphis, which has been key to multiplying training power and improving results.
One of the great novelties is the existence of two variants: Grok 4 Standard and Grok 4 Heavy. The Heavy version introduces a multi-agent model, where different agents work simultaneously on the same problem and the answers are compared to select the most appropriate solution. This version, aimed at professional users and companies, has early access to exclusive improvements and functionalities, although its price is significantly higher, around $300 per month.
Both versions, however, can be used both in chatbot format and through API, which facilitates their integration into third-party applications and projects. programming Advanced. For those looking for a specific tool for software development, xAI has launched Grok 4 Code, a variant designed to write, debug and optimize code efficiently.
Facing the giants: Grok 4 and its benchmarks
To measure the improvement of Grok 4 compared to competitors such as GPT-4.5 (OpenAI), Claude 4 (Anthropic) or Gemini 1.5 (Google), xAI has used independent tests and academic benchmarks. In the "Final Exam of Humanity," Musk's AI managed to solve around 25% of the questions—a figure higher than that of many rivals—and in tests such as MMLU-Pro, GPQA Diamond, or LiveCodeBench, It has outperformed competing models with success rates of up to 88%. in some cases.
The Heavy variant, as expected, is the one that performs best, especially in tests of advanced math, logic, and coding. However, experts have noted that These figures should be analyzed with caution until there is massive validation in real-world contexts, not just at the API or lab level.
Grok 4 also stands out for its real-time access to information through the social network XUnlike models that work only with trained, static data, you can incorporate current events into your responses, making it easier to verify, analyze trends, or instantly query news.
Controversies and adjustments: the challenge of moderation in AI
The launch of the Grok 4 has not been without controversy. Days before the official presentation, several responses to previous versions of the model were classified as anti-Semitic or offensive, provoking a reaction from both the tech community and X users themselves. xAI was forced to filter and remove inappropriate content, and has since communicated that The system has been retrained to reduce these risks and increase the security of responses.
Elon Musk himself has acknowledged that, in his attempt to neutralize biases and prevent manipulation, some temporary updates made the model "too complacent." After these setbacks, the idea is finding a balance between precision, creativity and responsibility, something that remains under scrutiny throughout the industry Generative AI.
Beyond the controversies, the regulatory context and government pressures also influence the evolution of Grok 4. The investment in this development frenzy is not minor: it is estimated that xAI spends about $1.000 billion a month in its operations, a sign of how tight the race for AI leadership is.
Price, access and first impressions of the market
Grok 4 is now available for those who have subscriptions to SuperGrok or SuperGrok Heavy, with annual prices of 255 euros and up to 2.556,99 euros respectively, although There are more affordable plans (around $30/month) for the standard versionX Premium+ users also have free access, a strategic move to gain market share against rivals that only offer demos or trial periods.
First impressions suggest that Grok 4 is shaping up to be a robust tool for those seeking advanced assistance with writing, analysis, or programming tasks, although it remains to be seen how it performs in everyday uses outside of a professional or academic environment.
Passionate writer about the world of bytes and technology in general. I love sharing my knowledge through writing, and that's what I'll do on this blog, show you all the most interesting things about gadgets, software, hardware, tech trends, and more. My goal is to help you navigate the digital world in a simple and entertaining way.