Hajra Naz
Elon Musk’s AI company, xAI, launched its latest flagship AI model, Grok 3, late Monday, along with new features for the Grok iOS and web apps.
Grok, designed as a competitor to models like OpenAI’s GPT-4 and Google’s Gemini, is capable of analyzing images and answering questions. It also powers several features on Musk’s social network, X. Although Grok 3 has been in development for months and was initially expected to be released in 2024, it missed that deadline.
The launch marks an ambitious milestone for xAI. The company has leveraged a massive data center in Memphis, home to roughly 200,000 GPUs, to train Grok 3. In a post on X, Musk revealed that Grok 3 was built with “10x” more computing power than Grok 2, utilizing an expanded training data set that reportedly includes court case filings.
Image Credits:xAI
“Grok 3 is an order of magnitude more capable than Grok 2,”
Musk stated during a live-streamed presentation on Monday.
“It’s a maximally truth-seeking AI, even if that truth sometimes goes against what is politically correct.”
Grok 3 is a family of models. The smaller version, Grok 3 mini, offers quicker response times at the expense of some accuracy. While not all features and models of Grok 3 are fully available yet (with some still in beta), they began rolling out on Monday.
According to xAI, Grok 3 outperforms GPT-4 on benchmarks such as AIME, which tests a model’s ability to solve math problems, and GPQA, which evaluates performance on PhD-level physics, biology, and chemistry questions. An early version of Grok 3 also performed strongly in Chatbot Arena, a crowdsourced platform where users vote on preferred responses from competing AI models, xAI claims.
Read More: Grok chatbot is made available to premium users by X
Similar to reasoning models like OpenAI’s o3-mini and DeepSeek’s R1 from the Chinese AI startup DeepSeek, two models in the new Grok 3 family—Grok 3 Reasoning and Grok 3 mini Reasoning—are made to systematically “think through” situations. These models aim to fact-check their responses before generating results, reducing common errors that typically affect AI models.
According to xAI, Grok 3 Reasoning outperforms OpenAI’s top o3-mini variant, o3-mini-high, on multiple benchmarks, including AIME 2025, a newly introduced mathematics evaluation.
These reasoning models are available through the Grok app, where users can ask Grok 3 to “Think” or use “Big Brain” mode for complex queries, which leverages additional computing power. xAI positions these models as ideal for mathematics, science, and programming-related questions.
Musk noted that some of the models’ reasoning processes are deliberately obscured in the Grok app to prevent distillation—a technique used by AI developers to extract knowledge from other models. This comes amid recent allegations that DeepSeek distilled OpenAI’s models to build its own.
Grok’s reasoning models also power DeepSearch, a new feature in the Grok app designed to compete with AI research tools like OpenAI’s deep research. DeepSearch scans the internet and X to analyze information and generate concise abstracts in response to user queries.
Subscribers to X’s Premium+ tier ($50 per month) will get early access to Grok 3, while additional features will be locked behind a new plan called SuperGrok. Priced at $30 per month or $300 annually (according to leaks), SuperGrok offers extended reasoning capabilities, additional DeepSearch queries, and unlimited image generation.
Shortly—possibly within a week—the Grok app will introduce a “voice mode,” allowing Grok models to generate synthesized speech, Musk revealed. A few weeks later, xAI plans to roll out Grok 3 models via its enterprise API, along with DeepSearch functionality.
Read More: X’s Grok AI chatbot is now available to all users
Musk also confirmed that xAI will open-source Grok 2 in the coming months.
“Our general approach is to open-source the previous version once the next version is fully released,” he explained.
“When Grok 3 reaches maturity and stability—likely in a few months—we’ll open-source Grok 2.”
When Musk first introduced Grok about two years ago, he positioned it as an AI model that was unfiltered, bold, and resistant to “woke” censorship—willing to tackle controversial topics that other AI systems avoided. To some extent, Grok and Grok 2 lived up to that claim, even generating explicit language when prompted—something models like ChatGPT typically avoid.
However, earlier versions of Grok often hedged on political issues, refusing to engage in certain discussions. One study even found that Grok leaned left on topics such as transgender rights, diversity initiatives, and economic inequality.
Musk attributed this to the model’s training data—public web pages—and vowed to make Grok more politically neutral. Whether xAI has succeeded in this effort remains uncertain, as do the potential implications of such changes.
