Elon Musk’s artificial intelligence startup, xAI, has introduced its latest model, Grok 3, claiming it surpasses rival offerings from OpenAI and China’s DeepSeek in early testing.

The model, tested on standardized math, science, and coding benchmarks, is being positioned as a major leap in AI reasoning 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,” Musk said during a demonstration streamed on his social media platform, X. Alongside Grok 3, xAI also announced Deep Search, a new product billed as a next-generation search engine.

The AI model will be rolled out later today for premium X subscribers and will also be available through a separate subscription for web and mobile access, according to xAI’s announcement.

READ MORE: OpenAI considers special voting rights to guard against takeovers

Musk Calls Grok 3 ‘Scary Smart’

Speaking at the World Governments Summit in Dubai last week, Musk described Grok 3 as “scary smart,” emphasizing its reasoning capabilities and asserting that it outperformed existing models in xAI’s internal tests.

He noted that the model was trained on “a lot of synthetic data” and designed to reflect on its mistakes to achieve greater logical consistency.

“This might be the last time that an AI is better than Grok,” Musk said, implying that future iterations will continue to push the boundaries of AI performance.

An early version of Grok 3 reportedly received higher ratings than competing models on Chatbot Arena, a crowdsourced platform that conducts blind tests between different AI systems.

xAI Expands GPU Power to Train Grok 3

xAI revealed that it doubled the size of its GPU cluster to train Grok 3, building on its existing Colossus supercomputer, which last year ran on a cluster of 100,000 advanced Nvidia GPUs.

The expansion highlights the company’s ambition to compete with AI giants like OpenAI while pushing the limits of AI model training capabilities.

Musk acknowledged that Grok 3 is still in beta, cautioning users to expect some imperfections initially, but promised rapid improvements. “We will improve it rapidly, almost every day,” he said, adding that voice assistance for the model would be introduced at a later stage.

Musk’s Growing Feud With OpenAI

Musk co-founded OpenAI with Sam Altman in 2015 as a nonprofit but has since become one of its biggest critics. Last month, he led an investor group that offered $97.4 billion to buy OpenAI’s nonprofit parent company—an offer OpenAI rejected.

Meanwhile, OpenAI remains a dominant player in the AI industry, having released its most advanced model, GPT-4o, in late 2023, with powerful reasoning abilities in science, math, and coding.

DeepSeek’s Rise Adds to AI Competition

The AI race intensified last month when Chinese start-up DeepSeek released research suggesting that one of its open-source models rivaled OpenAI’s GPT-4o while using less computing power.

This was particularly notable as the U.S. has restricted AI chipmaker Nvidia from selling its most advanced GPUs to China, limiting resources for Chinese firms developing large-scale AI models.

While some experts argue that DeepSeek’s breakthrough showcases AI innovation with fewer resources, others remain skeptical of its long-term impact.

With xAI scaling up its AI infrastructure and Grok 3 entering the market, competition among AI giants continues to heat up.

‍