AI news
February 18, 2025

Grok 3 Is Officially Here

Elon Musk and xAI team officially announce Grok 3 AI model on a live-streamed presentation.

Jim Clyde Monge
by 
Jim Clyde Monge

It’s official. Grok 3 is here.

On Monday, Elon Musk himself, along with three other xAI members, joined together for a livestreamed presentation of Grok 3.

If it’s your first time hearing of Grok, it’s an AI model developed by xAI to rival OpenAI’s GPT, Google’s Gemini, and the recently launched DeepSeek models.

Musk shared that the word ‘Grok’ came from Robert Heinlein’s sci-fi novel Stranger in a Strange Land. In the book, ‘Grok’ is a term used by a character raised on Mars, meaning to fully and deeply understand something.

You can watch the full livestreamed announcement here.

Members of the xAI team, including Musk (far right), during a live-streamed presentation of Grok 3.
Members of the xAI team, including Musk (far right), during a live-streamed presentation of Grok 3.

Days prior to the launch, Musk dubbed Grok 3 as the “smartest AI on earth.” During the live-streamed presentation, he added, “[It’s a] maximally truth-seeking AI, even if that truth is sometimes at odds with what is politically correct.”

xAI says Grok 3 is 10 to 15 times more powerful than Grok 2. It runs on the Colossus supercomputer, using 100,000 Nvidia H100 GPUs and racking up 200 million GPU-hours for training. With that kind of power, Grok 3 can handle massive datasets fast and accurately, pushing AI computing to a whole new level.

Grok 3 ranks number one on LMSYS’ Chatbot Arena with a big gap and scores impressively on pretraining and reasoning evaluations.

Grok 3 currently has over 1400 ELO score, with Gemini Flash Thinking coming in at second place with 1,385 ELO score.

It’s important to highlight the fact that Grok 3 is the first model ever to score over 1400 on Chatbot Arena and outperforms the best publicly available reasoning models from OpenAI and Google.

Grok 3 Benchmarks

Based on the benchmarks shared by xAI, Grok 3 beats GPT-4o on several comparisons, including AIME (which evaluates a model’s performance on a sampling of math questions) and GPQA (which assesses models using PhD-level physics, biology, and chemistry problems).

Grok 3 Benchmarks Based on the benchmarks shared by xAI, Grok 3 beats GPT-4o on several comparisons, including AIME (which evaluates a model’s performance on a sampling of math questions) and GPQA (which assesses models using PhD-level physics, biology, and chemistry problems).
Image from xAI

17 months after the original Grok model struggled with high school-level problems, Musk highlighted its rapid progress, stating that “Grok is ready to go to college” with how much it has improved.

Reasoning and a Mini Model

It’s worth noting that Grok 3 comes with a reasoning and a mini model.

  • Grok 3
  • Grok 3 Mini
  • Grok 3 Reasoning
  • Grok 3 Mini Reasoning

So basically, there are four variations of the Grok 3 model.

Grok 3 Mini can respond to questions more quickly at the cost of some accuracy. Not all the models and related features of Grok 3 are available yet (some are in beta), but they began rolling out on Monday.

Grok 3 Reasoning and Grok 3 Mini Reasoning can carefully “think through” problems, similar to “reasoning” models like OpenAI’s o3-mini, DeepSeek’s R-1, and Gemini-2 Flash Thinking.

Grok 3 Reasoning and Grok 3 Mini Reasoning can carefully “think through” problems, similar to “reasoning” models like OpenAI’s o3-mini, DeepSeek’s R-1, and Gemini-2 Flash Thinking.
Image from xAI

One impressive capability xAI showcased during the presentation is Grok 3’s ability to build games. In a demo, the team demonstrated how the model created a game blending elements of Tetris and Bejeweled.

One impressive capability xAI showcased during the presentation is Grok 3’s ability to build games. In a demo, the team demonstrated how the model created a game blending elements of Tetris and Bejeweled.
Image from xAI

This is an interesting capability from Grok considering that xAI is in plans to start an AI game studio. Musk retweeted the post below with the caption "yes,” confirming the news.

The focus is to build fun, engaging games with cutting-edge AI, avoiding political messaging, and challenging industry giants.

Progress at Ludicrous Speed

xAI has released new data showing Grok’s rapid advancement in language reasoning and computing power since 2023.

The graph below shows two things:

  1. Grok 2’s capabilities now surpass GPT-4’s benchmarks after just 18 months of development.
  2. Grok had a steep trajectory compared to OpenAI GPT’s gradual climb from 2019–2024.
Image from xAI

This is a bold and interesting claim, but the methodology on how these data were taken was not revealed, so it’s not clear how Grok was evaluated against GPT.

Grok 3 Availability

Grok 3 is now rolling out to Premium+ subscribers on X, which costs $50 per month. For those looking for even more advanced features, xAI is introducing a new subscription tier called SuperGrok, offering enhanced access to the AI model and additional capabilities.

The model is currently available through the Grok iOS app and the new Grok.com website, with plans to launch the app on Google Play soon. However, I checked the new website, and Grok 3 is currently not listed in the model dropdown option.

Image by Jim Clyde Monge

I suppose it’ll be available in the coming days.

The xAI team also confirmed that Grok 3 will be accessible via its enterprise API in a few weeks, alongside a feature called DeepSearch, which enhances search capabilities.

Musk announced that a voice mode will be added to the Grok app in about a week, allowing users to interact with the model using a synthesized voice. While Grok 3 is still considered a beta release, xAI encourages users to be mindful of potential errors as the model continues to improve.

Additionally, xAI reaffirmed its open-source strategy, stating that Grok 2 will be made open source once Grok 3 reaches a mature and stable stage, which is expected within a few months.

Get your brand or product featured on Jim Monge's audience