Elon Musk’s AI company, xAI, has unveiled its latest flagship AI model, Grok 3, alongside new features for the Grok iOS and web applications. Grok 3, developed to compete with models like OpenAI’s GPT-4 and Google’s Gemini, offers advanced capabilities such as image analysis and responsive answers to various queries. Although originally planned for release in 2024, Grok 3 has just launched after several months of development.
To train Grok 3, xAI has utilized a massive data center in Memphis, equipped with around 200,000 GPUs. Musk revealed that the new model has been developed with 10 times the computing power compared to its predecessor, Grok 2, using a broader dataset that may include court filings. During a live-streamed presentation, Musk highlighted Grok 3’s enhanced abilities, describing it as a “maximally truth-seeking AI,” even if that truth sometimes conflicts with political correctness.
The Grok 3 family includes several models, with Grok 3 mini being a smaller version that offers quicker responses at the expense of some accuracy. Not all of Grok 3’s features are available yet, but they began rolling out on Monday. xAI claims Grok 3 outperforms GPT-4 in benchmarks like AIME (which tests math problem-solving) and GPQA (which evaluates responses to PhD-level physics, biology, and chemistry problems). Grok 3 has also performed well in the Chatbot Arena, a crowdsourced test that pits different AI models against each other.
Among the new models is Grok 3 Reasoning and Grok 3 mini Reasoning, which excel at “thinking through” problems. These reasoning models, which are similar to OpenAI’s o3-mini and DeepSeek’s R1, prioritize fact-checking before providing answers. xAI claims Grok 3 Reasoning surpasses the best version of o3-mini in popular benchmarks, including AIME 2025. Users can access these reasoning models via the Grok app, where they can prompt the model to “Think,” or use “Big Brain” mode for more complex queries.
A new feature, DeepSearch, is also part of the Grok app, using reasoning models to scan the internet and X (formerly Twitter) to provide detailed summaries in response to user questions. DeepSearch competes with similar AI-powered research tools like OpenAI’s deep research. Subscribers to X’s Premium+ tier, priced at $50 per month, will get first access to Grok 3. Another subscription tier, SuperGrok, priced at $30 per month or $300 per year, will unlock additional features such as reasoning capabilities, DeepSearch queries, and unlimited image generation.
Looking ahead, Musk revealed that the Grok app will soon feature a “voice mode,” giving Grok models a synthesized voice. In the following weeks, Grok 3 models will also be available through xAI’s enterprise API. Musk also announced plans to open-source Grok 2 in the coming months, following the full release of Grok 3.
When Musk first introduced Grok two years ago, he positioned it as an AI model that would break boundaries, offering unfiltered and controversial responses that other AI systems might avoid. While Grok and Grok 2 fulfilled this promise in areas like vulgarity, they were still cautious on political topics. Musk has stated that Grok’s previous behavior was influenced by its training data and has committed to making Grok more politically neutral in the future. It remains to be seen whether this goal has been achieved with Grok 3 and what the implications of such neutrality might be.
