Elon Musk’s AI company, xAI, has launched its latest AI model, Grok 3, along with new features for its iOS and web apps.
Grok, xAI’s response to OpenAI’s GPT-4o and Google’s Gemini, can analyze images and answer questions. It also integrates with Musk’s social network, X. Grok 3 has been in development for months and was initially set for release in 2024 but missed the deadline, according to TechCrunch.
Vast training dataset
XAI trained Grok 3 using a massive data center in Memphis, equipped with around 200,000 GPUs. In a post on X, Musk claimed that Grok 3 was built with ten times more computing power than its predecessor, Grok 2. Additionally, it was trained on a more extensive dataset, including legal documents.
Not all features of Grok 3 are available yet, as some are still in beta. However, the rollout has begun. According to Musk, Grok 3 is significantly more capable than Grok 2. He described it as an AI that seeks truth to the maximum, even when that truth challenges political correctness.
Performance and benchmarks
XAI claims that Grok 3 surpasses GPT-4o in key benchmarks. These include AIME, which tests a model’s math skills, and GPQA, which evaluates knowledge in PhD-level physics, biology, and chemistry. An early version of Grok 3 also performed well in Chatbot Arena, a user-managed test where AI models compete, and users vote for the best answers.
Two new reasoning models
Grok 3 is a family of models. A smaller version, Grok 3 mini, offers faster responses at the cost of some accuracy. Additionally, two new reasoning models—Grok 3 Reasoning and Grok 3 mini Reasoning—enhance problem-solving abilities. These models function similarly to OpenAI’s o3-mini and DeepSeek’s R1, as they attempt to verify their answers before providing them, reducing common AI errors.
Users can access these reasoning models through the Grok app. They can also enable “Big Brain” mode, which utilizes extra computational power for complex queries. xAI states that these models are particularly effective for math, science, and programming questions.
Musk mentioned that some reasoning model insights are hidden in the Grok app to prevent distillation, a technique AI developers use to extract knowledge from other models. Recently, DeepSeek was accused of using this method on OpenAI’s models.
New features and pricing plans
The reasoning models support a new feature in the Grok app called DeepSearch. This tool, xAI’s answer to AI-driven research tools like OpenAI’s deep research, scans the Internet and X to analyze information and summarize responses.
X’s Premium+ subscribers ($50 per month) will be the first to access Grok 3. Other features will be available through a new subscription plan called SuperGrok. Leaked information suggests that SuperGrok costs $30 per month or $300 per year and provides additional reasoning features, DeepSearch queries, and unlimited image generation.
Musk also revealed that the Grok app will gain a voice mode within a week, allowing Grok models to use a synthesized voice. A few weeks later, Grok 3 will be accessible via xAI’s business API, along with DeepSearch functionality. Additionally, xAI plans to open-source Grok 2 in the coming months.
Political bias and controversies
When Musk introduced Grok two years ago, he positioned it as an AI that was “edgy, unfiltered, and anti-woke,” capable of addressing controversial topics that other models avoid. In some cases, this has been true. When asked to use explicit language, Grok and Grok 2 complied without hesitation, unlike models such as ChatGPT.
However, previous Grok models remained cautious on political issues, avoiding certain topics. A study even found that Grok exhibited a left-wing bias on subjects like transgender rights, diversity programs, and inequality.
Musk attributed this bias to Grok’s training data, which primarily consists of publicly available web pages. He pledged to make Grok more politically neutral. Whether xAI has achieved this goal remains unclear, and its implications are yet to be seen.
