Saturday, October 18, 2025
Home Innovation Artificial Intelligence Elon Musk's xAI Introduces Gro...
Artificial Intelligence
Business Honor
20 Febuary, 2025
xAI's Grok-3 outperforms rivals, introduces Deep Search, and claims to seek controversial truths.
Elon Musk's AI firm, xAI, has released a new iteration of its artificial intelligence model named Grok-3. Musk declares that Grok-3 is significantly more capable than its earlier version, Grok-2, and that it surpasses leading AI models from U.S. and Chinese companies.
Grok-3 is coming out to X’s Premium+ subscribers first, and also a new, more expensive subscription called SuperGrok for access to more advanced features. Musk says that Grok-3 is "an order of magnitude more capable" than before. It's still in development, but users can look for regular improvements, such as a voice interaction feature soon.
The developers say Grok-3 was trained with 10 times more computing power than Grok-2 and has outperformed other AI models in tasks like math, science, and coding. For example, it beat models like DeepSeek-V3 and GPT-4o on these tasks. However, it still lags behind a few models like DeepSeek-R1 and OpenAI’s models.
In addition to that, xAI launched a product known as Deep Search, which is a next-gen search engine. It is an instrument capable of browsing web pages and X posts efficiently to provide answers, making research time-consuming.
Musk also characterized Grok as an AI that is tasked to "seek the truth," whether it's argumentative or politically incorrect. The model's bold or provocative statements have given rise to the theory that Musk is framing its opinions, especially when it panned certain media such as The Information while touting X as the repository of "raw, unfiltered news."
However, NBC News's numerous tests of Grok-3 on Tuesday revealed that the model failed to generate such a response. The Information is a “well-regarded tech news outlet known for its in-depth reporting and analysis,” according to Grok-3, who frequently stated that the magazine is “not infallible” due to its paywalled content and specialized emphasis when posed the same question.