TLDR
- Elon Musk’s xAI launched Grok 3, powered by a massive 200,000 GPU cluster called Colossus, claiming superior performance over competitors in math, science, and coding tests
- The model will be available to X Premium subscribers and through a separate subscription, with additional features like DeepSearch and voice assistance coming soon
- Early testing shows Grok 3 outperforming rivals on Chatbot Arena, with its Reasoning Beta variant achieving 93% on AIME 2025 benchmark
- xAI plans to expand Colossus to five times its current capacity and will open-source Grok 2 in the coming months
- The company announced plans for an AI gaming studio and API access for developers, while demonstrating capabilities in physics problem-solving and game code generation
Elon Musk’s artificial intelligence company xAI unveiled its latest AI model, Grok 3, on Tuesday, February 18, 2025. The release showcases the company’s expanded computing infrastructure and claims improved performance across various benchmarks.
At the heart of Grok 3’s development is Colossus, a supercomputer cluster utilizing 200,000 GPUs for AI training. The system was built in two phases, with initial training taking 122 days on 100,000 GPUs, followed by a 92-day scaling period to reach full capacity.
— xAI (@xai) February 18, 2025
The model will be available to X Premium subscribers starting Tuesday in the United States, with access also provided through a separate subscription for web and app versions. XAI announced plans to introduce voice assistance features in the coming weeks.
During a live demonstration streamed on X, Musk expressed enthusiasm about the new model’s capabilities. “We’re very excited to present Grok 3, which is, we think, an order of magnitude more capable than Grok 2 in a very short period of time,” he said.
Early testing results show promising performance. The base model topped charts in math (AIME), science (GPOA), and coding (LCB) tests. A specialized version called Reasoning Beta achieved 93% on the AIME 2025 benchmark, surpassing competing models that scored below 87%.
The company revealed that an early version of Grok 3, codenamed “Chocolate,” received the highest ratings on Chatbot Arena, a platform where users evaluate AI models without knowing their identity. This blind testing approach helps prevent potential bias in evaluation.
200,000 GPUs Power xAI’s Latest Release
XAI introduced a new feature called DeepSearch, described as a “next generation search engine.” The tool is designed to search the web and generate comprehensive reports on various topics, similar to existing offerings from other AI companies.
The development team demonstrated Grok 3’s abilities in solving physics problems and writing game code. They also announced plans for an AI gaming studio that will allow developers to create games powered by the new model.
Some early users have shared positive experiences. Computer scientist Lex Friedman praised the model’s capabilities, while X user Penny2x demonstrated Grok 3’s ability to create a 2D platformer game through iterative improvements.
Former OpenAI co-founder Andrej Karpathy compared Grok 3’s performance to existing models, stating it “feels somewhere around the state of art territory of OpenAI’s strongest models.”
I was given early access to Grok 3 earlier today, making me I think one of the first few who could run a quick vibe check.
Thinking
✅ First, Grok 3 clearly has an around state of the art thinking model (“Think” button) and did great out of the box on my Settler’s of Catan… pic.twitter.com/qIrUAN1IfD— Andrej Karpathy (@karpathy) February 18, 2025
XAI confirmed plans to open-source Grok 2 once Grok 3 is fully operational, continuing their practice of releasing previous versions to support innovation in the field.
The company plans to expand the Colossus supercomputer to five times its current capacity, which would make it the most powerful GPU cluster globally.
API access for developers will be released in the upcoming weeks, along with audio transcription capabilities, enabling third-party applications to utilize Grok 3’s features.
The rollout of Grok 3 is happening gradually, with full access and additional features expected to become available in the coming weeks.
Musk had previously described the model as “scary smart” during the World Governments Summit in Dubai, claiming it could reflect on its mistakes to achieve logical consistency.
