Late Monday evening, Elon Musk’s AI venture, xAI, unveiled its newest AI flagship model, Grok 3, along with enhanced features for the Grok application available on iOS and the web.
Grok serves as xAI’s competitor to other leading models such as OpenAI’s GPT-4o and Google’s Gemini, boasting capabilities that include image analysis and question answering. It also powers numerous functionalities on Musk’s social platform, X. Although Grok 3 had been in progress for several months and was originally expected to debut in 2024, it ultimately missed that timeline.
To train Grok 3, xAI has utilized a massive data center in Memphis, hosting approximately 200,000 GPUs. In a tweet on X, Musk claimed that Grok 3 uses “10x” more computational power than its predecessor, Grok 2, and employs a more extensive training dataset that reportedly includes legal filings.
“Grok 3 is significantly more powerful than Grok 2,” Musk stated during a live presentation on Monday. “[It’s a] maximally truth-seeking AI, even if that truth sometimes challenges political correctness.”
The Grok 3 model family consists of multiple variants — not limited to just one. A compact iteration, Grok 3 mini, offers quicker responses, though this speed comes at the expense of some accuracy. While not all models may be accessible just yet, the rollout commenced on Monday.
According to xAI, Grok 3 outperforms GPT-4o on various evaluations, including AIME, which assesses mathematical ability, and GPQA, which tests knowledge in physics, biology, and chemistry at the doctorate level. Furthermore, an early iteration of Grok 3 performed competitively in Chatbot Arena, a user-driven assessment platform where different AI models are tested and compared, as reported by xAI.
Two specific variations — Grok 3 Reasoning and Grok 3 mini Reasoning — are designed to methodically assess problems, akin to reasoning models such as OpenAI’s o3-mini and China’s DeepSeek’s R1. These reasoning models prioritize self-fact-checking before producing outcomes to circumvent common errors that typically affect AI models.
xAI asserts that Grok 3 Reasoning outperforms the top-performing version of o3-mini — o3-mini high — across several established benchmarks, including the updated mathematics standard, AIME 2025.
Users can access the reasoning capabilities of Grok through the Grok app, allowing them to prompt Grok 3 to engage in deeper thinking for more complex inquiries by using “Big Brain” mode. This mode is particularly beneficial for resolving questions related to mathematics, science, and coding.
Musk mentioned that some elements of the reasoning process are being kept hidden to avert distillation, a practice whereby AI developers extract knowledge from one model to enhance another. Recently, Chinese firm DeepSeek faced accusations for distilling information from OpenAI’s models to create its AI solutions.
Grok’s reasoning mode adds to another newly introduced function called DeepSearch. This feature serves as xAI’s response to AI-driven “deep research” tools like OpenAI’s Deep Research. DeepSearch scours the internet and the X platform to gather and synthesize information, providing concise summaries in reply to user inquiries.
Grok 3 will initially be available to subscribers of X’s Premium+ plan, with additional features accessible through xAI’s SuperGrok subscription, priced at $30 each month or $300 annually. This includes access to extended reasoning capabilities, DeepSearch queries, and unlimited image generation.
Musk indicated that Grok will soon introduce a voice mode, anticipated to launch within about a week. Subsequently, Grok 3 models will be integrated into xAI’s API within a few weeks thereafter.
When Musk initially revealed Grok nearly two years ago, he positioned the AI as a bold, unfiltered alternative, one willing to tackle contentious topics that other AI systems might avoid. In certain instances, he delivered on that promise; for instance, Grok and Grok 2 had no problem answering with explicit language when prompted.
However, the earlier Grok iterations were known to tread carefully around political contexts, often refraining from addressing certain sensitive issues. Studies have indicated that Grok tended to lean left on topics like transgender rights and diversity initiatives.
Musk has attributed these tendencies to the training data utilized — specifically, publicly available web content — and has committed to making Grok more politically neutral. It remains to be seen if xAI has succeeded in achieving that objective.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
