Elon Musk announces Grok 4 as the world’s most powerful AI during live launch

The company xAI, founded by Elon Musk, officially launched the Grok 4 on Tuesday night (9), during a live broadcast. The model was presented as the most powerful artificial intelligence in the world to date. The tool outperformed competitors such as Gemini 2.5 Pro, Claude 4 Opus and OpenAI’s o3 in several independent benchmarks.
The announcement took place at 11pm (Pacífico time) and was attended by Elon Musk and engineers from the team. The company highlighted that the Grok 4 was trained with its own infrastructure and presents significant advances in logical reasoning and complex problem solving.
Performance in main tests
Grok 4 recorded 88% correct answers on the GPQA Diamond exam, which assesses PhD level knowledge. The result surpasses the previous record of 84% for Gemini 2.5 Pro.
In the Humanity’s Last Exam test, the model achieved 44.4% in the heavy version, against 26.9% for the best previous competitor. The score in advanced mathematics reached 100% in AIME 2025 and 96.7% in Harvard-MIT Math.
- GPQA Diamond: 88%
- Humanity’s Last Exam (heavy): 44.4%
- AIME 2025: 100%
- Harvard-MIT Math: 96.7%
- LiveCodeBench (coding): 79.4%

Available versions and access
xAI made two variants of the model available. The standard Grok 4 serves common users with fast and accurate responses. Já version Grok 4 Heavy uses multiple simultaneous agents to solve more difficult questions.
Access to Grok 4 is initially restricted to Premium+ subscribers on X and SuperGrok on the grok.com website. The company reported that version Heavy requires a specific, higher-cost plan.
Users have reported response times of up to 75 seconds on complex issues with the heavy-duty variant. The tool demonstrated the ability to correct errors in questions provided during the presentation.
Highlighted technical capabilities
The model solves real engineering problems that don’t exist on the internet or in books. The team stated that Grok 4 identifies ambiguities in questions and provides complete answers.
In the demonstration, the system solved advanced physics and mathematics questions in real time. The company emphasized that the model represents a significant leap compared to the previously launched Grok 3.
Integration with platforms
Grok 4 is now available in the X app and on the website grok.com. Version Heavy only works on xAI’s dedicated portal for now.
The company plans to gradually expand access over the coming months. Usuários can test the standard model immediately after updating the app or subscribing to the Premium+ plan.
Next developments
xAI works on continuous improvement of Grok 4. The team mentioned that new imaging and video generation functionalities will be incorporated in future updates.
The heavy model will undergo refinements to reduce response time. The company remains focused on developing artificial intelligence aligned with scientific discovery.
Current availability
Grok 4 can be accessed immediately by eligible subscribers. The standard version answers general questions with high accuracy.
The Heavy variant is limited to complex issues and requires a specific signature. xAI recommends using the standard model for everyday queries.
The launch reinforces xAI’s position in the artificial intelligence market. The Grok 4 sets a new standard for performance in academic and practical tests.






