AI Models: Musk’s Grok 3 AI Model Drops; Meta to Hold LlamaCon

The artificial intelligence (AI) race continues to heat up with Elon Musk’s AI startup, xAI, unveiling the latest version of its AI model, Grok 3.

In an hour-long livestreamed event on X, the team said Grok 3 beats OpenAI’s GPT-4o, Google’s Gemini 2 Pro, DeepSeek’s V3, Anthropic’s Claude 3.5 Sonnet and even its own Grok 3-mini on math, science and coding benchmarks.

Training Grok 3 took double the computing power than Grok 2, the team said. In April 2024, xAI built its down data center with 100,000 GPUs. It was later expanded to 200,000 GPUs, which was used to train Grok 3. XAI used its custom supercomputer, Colossus, for the job.

Grok 3, which is both a reasoning and generalist model, is 10 times more powerful than Grok 2. It is available to subscribers of X’s Premium+ service, which costs $40 a month.

OpenAI Co-founder Andrej Karpathy, who has left the startup and founded Eureka Labs, was granted early access to test Grok 3. He said on a post on X that Grok 3 has similar capability as OpenAI’s o1-pro AI model ($200 a month) and is better than DeepSeek’s R1 model.

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

Karpathy also added that Grok’s “Deep Search” feature is on par with Perplexity’s “DeepResearch” offering but not yet at the level of OpenAI’s “Deep Research,” which creates research papers.

Meanwhile, OpenAI plans to release GPT-4.5 and GPT-5 in weeks or months. In a post on X, OpenAI CEO Sam Altman said GPT-4.5 has been “much more of a ‘feel the AGI’ [artificial general intelligence] moment among high-taste testers than I expected!”

trying GPT-4.5 has been much more of a “feel the AGI” moment among high-taste testers than i expected!

— Sam Altman (@sama) February 17, 2025

Microsoft, which has invested close to $14 billion in OpenAI, cannot use OpenAI’s AGI-level AI models for its Azure cloud platform, based on their licensing agreement.

But the Financial Times reported that OpenAI is considering removing that restriction from its corporate structure to attract more investments, since training AI models is a costly endeavor. The clause was included to keep the most powerful AI capabilities from wreaking unintended consequences.

Meta has announced LlamaCon, its first developer conference for its flagship Llama, a family of large language models that are free and open source. The conference will be held on April 29.

Meta said Llama is the most adopted AI model globally, with 650 million downloads, twice as many as three months ago. Demand for Llama is soaring, with license approvals more than doubling in the past six months, according to the company.

“The growth of Llama … was exponential this year,” Ahmad Al-Dahle, Meta’s vice president of generative AI, said in a blog post.

The latest publicly available Llama version is Llama 3, with Llama 4 potentially launching this year. “As we look to 2025, the pace of innovation will only increase as we work to make Llama the industry standard for building on AI,” Al-Dahle wrote.

He also said that the company’s AI assistant, Meta AI, is on track to be the world’s most-used assistant by the end of the year, with 600 million monthly active users. Meta AI uses Llama 3.

Meta AI is embedded in Meta’s social media and chat channels: Facebook, WhatsApp, Instagram and Messenger.

Mira Murati, the former CTO of OpenAI, on Tuesday (Feb 18) launched a new AI startup called Thinking Machines Lab that will focus on human-AI collaboration.

“Instead of focusing solely on making fully autonomous AI systems, we are excited to build multimodal systems that work with people collaboratively,” according to the startup’s blog post. “We see enormous potential for AI to help in every field of work.”

Murati, who is CEO of the startup, hopes to make AI more understandable and accessible to all. The scientific community lags in their understanding of the advancing capabilities of frontier AI models — defined as boundaries-pushing foundation AI models, according to the blog post.

This knowledge is “concentrated” in the top research labs, which “limits both the public discourse on AI and people’s ability to use AI effectively,” the startup said.

In addition, these AI models have plenty of potential but remain “difficult” for people to customize for their specific needs, the post said.

Murati left OpenAI last September to “create time and space to do my own exploration.” She was often the public-facing OpenAI executive that accompanied CEO Sam Altman to events and interviews.

Murati’s exit follows a slew of high-profile resignations at OpenAI, including Co-founder and Chief Scientist Ilya Sutskever, who launched his own startup, Safe Superintelligence Inc.

Former OpenAI colleagues that joined Murati’s startup include OpenAI Co-founder John Schulman, who is now chief scientist, and Barret Zoph, OpenAI’s vice president of post-training who is now CTO.

The post AI Models: Musk’s Grok 3 AI Model Drops; Meta to Hold LlamaCon appeared first on PYMNTS.com.