Ілон Маск і команда xAI / X
Elon Musk’s artificial intelligence company has released its latest flagship AI model, Grok 3 — xAI’s answer to OpenAI and China’s DeepSeek reasoning models, which is being promoted as «scary smart AI».
It is noted that to train Grok 3, xAI used a huge data center in Memphis, which contains about 200,000 GPUs. Musk himself noted that the latest model was created with «10 times» more computing resources than Grok 2 and with an expanded training data set that allegedly includes court case materials.
«Grok 3 is an order of magnitude more powerful than Grok 2», Musk said during a presentation broadcast on Monday. «It’s an AI that seeks the truth as much as possible, even if that truth sometimes contradicts what is politically correct».
Grok 3 — is not a single model, but a whole family, which also includes a smaller version of Grok 3 mini (it answers questions faster, but may not be as accurate). Currently, only certain models are available in beta, while the full launch is promised for Monday.
According to xAI, Grok 3 outperforms GPT-4o on tests — including AIME, which evaluates model performance on a sample of math questions, and GPQA, which evaluates models on PhD-level physics, biology, and chemistry problems. An early version of Grok 3 also allegedly achieved competitive results in the Chatbot Arena tests, which pits different AI models against each other and collects user ratings for their answers.
The two variations of Grok 3 — Grok 3 Reasoning and Grok 3 mini Reasoning — can carefully «think through» problems, similar to «reasoning models» such as o3-mini by OpenAI й R1 of the Chinese DeepSeek. Reasoning models are known for checking themselves thoroughly before publishing results, and this allows them to avoid some of the pitfalls that typically slow down «standard models.
Musk’s startup also boldly claims that Grok 3 Reasoning outperforms the best version of o3-mini — o3-mini-high — on several popular tests, including a newer math test called AIME 2025.
The reasoning models that xAI promotes as best suited for math, science, or programming queries will be available in the web version of Grok or the iOS app — for more complex tasks, there are options for «Think» and «Big Brain» (the latter involving «additional computation»), and there is also a DeepSearch option (a deep search like the one recently presented OpenAI).
Musk noted that some «thoughts» reasoning models in the program will be hidden to prevent distillation — a method currently used by AI model developers to borrow data from other models. OpenAI has recently suspected that this is how DeepSeek model was trained, which allegedly took data from ChatGPT.
Grok 3 will initially be available to X Premium+ subscription holders, but for advanced features, a new plan called SuperGrok is now offered, which for $30 per month (or $300 per year), unlocks additional DeepSearch reasoning and queries, and provides unlimited image creation.
In about a week, Grok 3 (and its predecessors) will receive «voice mode», and in a few weeks it will be included in the xAI enterprise API along with DeepSearch. Musk also promises to open source Grok 2 in a few months.
«Our approach is that we open source the latest version when the next version is fully released. When Grok 3 is mature and stable, which is probably within a few months, then we will open source Grok 2».
Source: TechCrunch