xAI's Grok 3 model excels in reasoning & challenges OpenAI's o-series

xAI announced Grok 3, which is significantly more powerful than its predecessor, Grok 2. With that said, the Elon Musk-led company is challenging OpenAI and its o-series models that are capable of reasoning.

Musk previously hyped Grok 3 as a “scary smart” model, and it backs it up by having exceptional reasoning capabilities. In xAI’s words, the model’s reasoning performance outperforms OpenAI’s o1 and DeepSeek-R1, the so-called cheap model that rocked the AI world not too long ago, when given more test-time computing.

The model’s capacity was doubled in just 92 days using 200,000 GPUs, and it underwent 10 times more training than Grok 2, finishing pretraining in early January but continuing to train.

Benchmark results show Grok 3 outperforms competitors like Gemini 2 Pro and GPT-4o, with even the smaller Grok-3 mini being competitive. In the Chatbot Arena (LMSYS), Grok 3 achieved an Elo score of 1400, the highest recorded so far, and its performance continues to improve.

Grok 3 is now accessible to all users on X (formerly Twitter). Initially exclusive to Premium subscribers, Grok 3 has been made available to free users, allowing them to send up to 10 messages every two hours.

You can also access the model directly through the X platform without the need for additional applications.

The news arrived amid OpenAI’s recent push for its upcoming model, GPT-4.5, which the company’s CEO Sam Altman hyped as the closest thing we have to artificial general intelligence (AGI).

Both GPT-4.5 and GPT-5 models are coming in a matter of “weeks or months,” and once they do, OpenAI will eliminate ChatGPT’s model picker to give everyone a one-stop simplified AI experience.