Elon Musk’s artificial intelligence firm, xAI, has rolled out Grok 3, the latest version of its AI model, alongside updates for its iOS and web applications. The launch follows months of development, with the model initially planned for a 2024 release but delayed beyond that timeframe. Grok, which powers features on Musk’s social platform X, is designed to compete with OpenAI’s GPT-4o and Google’s Gemini.
To develop this model, xAI relied on a vast computing network, tapping into a Memphis-based data center equipped with nearly 200,000 GPUs. Musk has stated that Grok 3 required ten times more computing resources than its predecessor. The training dataset was also significantly expanded, reportedly incorporating a broad set of documents, including legal records. Unlike a single AI system, Grok 3 consists of multiple versions, one of which, Grok 3 mini, delivers quicker responses by prioritizing speed over absolute accuracy. While some features are still in beta testing, the initial rollout began on Monday.
Performance evaluations suggest that Grok 3 surpasses OpenAI’s GPT-4o in various assessments. xAI claims the model demonstrates strong reasoning capabilities, particularly in solving mathematical and scientific problems. Early tests placed it ahead in competitive AI rankings, where users compare model-generated responses.
The latest release introduces enhanced reasoning models, including Grok 3 Reasoning and its smaller counterpart, which focus on refining logical analysis. These systems attempt to validate their own outputs before responding, aiming to reduce factual inconsistencies. According to xAI, Grok 3 Reasoning exceeds OpenAI’s o3-mini-high in several benchmarks, including mathematics assessments.
Users can engage these advanced features within the Grok application, where a mode called "Big Brain" allows for deeper computational analysis. These models are particularly geared toward technical disciplines, offering more precise assistance in programming, scientific research, and complex problem-solving. To prevent unauthorized knowledge extraction, xAI has also introduced safeguards that conceal certain internal processes. This measure follows industry concerns over distillation, a method where AI models learn from competing systems.
Another addition is DeepSearch, a tool designed to scan the internet and X’s platform to compile structured responses based on user queries. This feature serves as xAI’s alternative to OpenAI’s research-oriented tools, offering AI-powered content analysis.
Access to Grok 3 remains limited initially, with availability restricted to X’s Premium+ subscribers, who pay $50 per month. xAI is also preparing to launch a new tier, SuperGrok, which is expected to cost $30 per month or $300 annually. This subscription unlocks additional reasoning capabilities, expands DeepSearch queries, and offers unlimited image generation.
Musk has indicated that a voice interaction feature will soon be incorporated into the Grok application, allowing users to engage with the AI through speech. In the following weeks, xAI intends to release Grok 3 and DeepSearch to enterprise clients via an API.
As part of its broader strategy, xAI plans to open-source Grok 2 once Grok 3 is fully established. Musk has stated that prior versions will be publicly released when the newest iteration reaches maturity, though no exact timeline has been provided.
When Musk first introduced Grok, he framed it as an AI capable of addressing controversial topics more openly than its competitors. Previous versions, including Grok 2, displayed a more flexible conversational approach, even engaging in explicit language upon request. However, studies suggested that earlier models leaned toward progressive viewpoints on social and political matters, particularly regarding diversity programs and economic policies. Musk attributed this to biases within publicly available training material and has vowed to adjust the model’s ideological stance to achieve a more neutral position. Whether these changes have taken full effect in Grok 3 remains uncertain, leaving questions about their broader impact on user interactions.
Read next: Researchers Uncover YouTube’s True Scale as Google Withholds Platform Insights
To develop this model, xAI relied on a vast computing network, tapping into a Memphis-based data center equipped with nearly 200,000 GPUs. Musk has stated that Grok 3 required ten times more computing resources than its predecessor. The training dataset was also significantly expanded, reportedly incorporating a broad set of documents, including legal records. Unlike a single AI system, Grok 3 consists of multiple versions, one of which, Grok 3 mini, delivers quicker responses by prioritizing speed over absolute accuracy. While some features are still in beta testing, the initial rollout began on Monday.
Performance evaluations suggest that Grok 3 surpasses OpenAI’s GPT-4o in various assessments. xAI claims the model demonstrates strong reasoning capabilities, particularly in solving mathematical and scientific problems. Early tests placed it ahead in competitive AI rankings, where users compare model-generated responses.
The latest release introduces enhanced reasoning models, including Grok 3 Reasoning and its smaller counterpart, which focus on refining logical analysis. These systems attempt to validate their own outputs before responding, aiming to reduce factual inconsistencies. According to xAI, Grok 3 Reasoning exceeds OpenAI’s o3-mini-high in several benchmarks, including mathematics assessments.
Users can engage these advanced features within the Grok application, where a mode called "Big Brain" allows for deeper computational analysis. These models are particularly geared toward technical disciplines, offering more precise assistance in programming, scientific research, and complex problem-solving. To prevent unauthorized knowledge extraction, xAI has also introduced safeguards that conceal certain internal processes. This measure follows industry concerns over distillation, a method where AI models learn from competing systems.
Another addition is DeepSearch, a tool designed to scan the internet and X’s platform to compile structured responses based on user queries. This feature serves as xAI’s alternative to OpenAI’s research-oriented tools, offering AI-powered content analysis.
Access to Grok 3 remains limited initially, with availability restricted to X’s Premium+ subscribers, who pay $50 per month. xAI is also preparing to launch a new tier, SuperGrok, which is expected to cost $30 per month or $300 annually. This subscription unlocks additional reasoning capabilities, expands DeepSearch queries, and offers unlimited image generation.
Musk has indicated that a voice interaction feature will soon be incorporated into the Grok application, allowing users to engage with the AI through speech. In the following weeks, xAI intends to release Grok 3 and DeepSearch to enterprise clients via an API.
As part of its broader strategy, xAI plans to open-source Grok 2 once Grok 3 is fully established. Musk has stated that prior versions will be publicly released when the newest iteration reaches maturity, though no exact timeline has been provided.
When Musk first introduced Grok, he framed it as an AI capable of addressing controversial topics more openly than its competitors. Previous versions, including Grok 2, displayed a more flexible conversational approach, even engaging in explicit language upon request. However, studies suggested that earlier models leaned toward progressive viewpoints on social and political matters, particularly regarding diversity programs and economic policies. Musk attributed this to biases within publicly available training material and has vowed to adjust the model’s ideological stance to achieve a more neutral position. Whether these changes have taken full effect in Grok 3 remains uncertain, leaving questions about their broader impact on user interactions.
Read next: Researchers Uncover YouTube’s True Scale as Google Withholds Platform Insights
