SEARCH
SHARE IT
Elon Musk's xAI has officially unveiled Grok 3, dubbed "the smartest AI on Earth". On several benchmarks, the model matches, if not outperforms, top AI models such as OpenAI's GPT-4o and Google's Gemini 2.0 Flash Thinking.
The model was trained on over 100,000 Nvidia H100 GPUs, resulting in 200 million GPU hours, ten times more than was needed to train Grok 2.
Grok 3 outperforms GPT-4o in several benchmarks, including AIME (math) and GPQA (PhD-level physics, biology, and chemistry). It also performed well in Chatbot Arena, a crowdsourcing test. Grok 3's reasoning model outscored others in the AIME 2024 exam, as well as a smaller version of the fresh AIME 2025 exam.
The Grok 3 series features a reasoning model that "thinks" before answering, resulting in a more accurate response. Users can choose the "big brain" mode for more challenging questions, which xAi describes as best suited for maths, physics, and computing. Grok 3's reasoning mode is also combined with DeepSearch, which, as the name implies, searches the internet and X for information and provides extensive responses to requests. There is also a Grok 3 mini that might be utilised to produce faster results.
The reasoning models are accessible through Grok's standalone app, though Musk stated that portions of the model's "thoughts" are concealed to prevent distillation, in which another "child" AI model can extract knowledge from its "parent" model. DeepSeek was recently accused of a similar offence, with the investigation focussing on whether it distilled knowledge from OpenAI's model to train its own.
Grok 3 will be available first to X's Premium+ members for $22 per month, with additional features reserved for a separate forthcoming subscription called SuperGrok, which will enable access to reasoning capabilities, DeepSearch queries, and infinite image production. Musk also stated that Grok will soon have a voice option, and that the Grok 3 models will be available through xAI's API in a few weeks.
MORE NEWS FOR YOU