Grok 3 by xAI 🤯

the most capable AI model that EVER existed

Elon and the xAI team have released the strongest AI model that EVER existed.

I know you heard this already many times, but this is how AI evolves. There are better versions every now and then. However, this time, we are speaking about a truly remarkable model. The first that ever hit an ELO score above 1400 and rising!

In this episode: Some early details of it, how to get access, and what I think after using it a couple times! (I already have access!✨)

The Details

Grok 3 is the first model to score over 1400 on Chatbot Arena and outperforms the best publicly available reasoning models from OpenAI and Google.

xAI was founded 13 years after Deepmind and 8 years after OpenAI and is now ahead of both.

Its ELO score is 1402 to be precise.

A table from Imarena.ai displaying Chatbot Arena rankings, highlighting xAI’s Grok 3 as the #1 model with an Arena Score of 1402. The table lists models, their scores, 95% confidence intervals, vote counts, and organizations. Grok 3, labeled as 'chocolate... (early_grok-3),' is marked in green, indicating its top position. Other models like Gemini 2.0, GPT-4o, and DeepSeek-R1 follow, with scores ranging from 1385 to 1305. The background is dark with white text, and the title reads 'xAI Grok-3 #1 in Arena, First model with >1400 score' in green and white text.
A table from Imarena.ai showing Grok 3’s performance across various categories, with the title 'Grok-3 is also #1 across all categories' in large white text on a dark background. The table lists models and their rankings in categories like Overall, Hard Prompts, Coding, Math, Creative Writing, Instruction Following, Longer Query, and Multi-Turn. Grok 3, labeled as 'early_grok-3,' is ranked #1 in every category, highlighted in yellow. Other models like GPT-4o, Gemini 2.0, and DeepSeek-R1 follow, with rankings ranging from 2 to 18. The background is dark with white and light text for readability.

How did they get to this performance?

Among many technical details, two facts stand out. First, Grok 3 was trained on 10x more compute than Grok 2 (Grok 2 equivalent to GPT-4o). Grok 3 was trained on xAI’s “Colossus” supercomputer, leveraging 100,000 Nvidia H100 GPUs, with reports suggesting it consumed around 200 million GPU hours.

Second, the focus on logical consistency through mostly synthetic data. In the demo, Musk has emphasized its use to achieve "logical consistency" and reduce reliance on potentially messy or legally contentious internet-scraped data.

How do you get access?

Not that complicated, actually. Only prerequisite, you need to have a Premium+ account. (~20$/m)

There is also https://grok.com, where you get a more ChatGPT-ish work environment. However, it is currently not accessible in the EU. (đź’€)

My insights, working with the model

I think it is fast, and very intelligent. It feels different, one can feel the intelligence, perhaps the AGI. It is hard for me to describe it, but while it is slightly more direct (it doesn’t want you to desperately like it) and less politically correct, the interactions are very insightful.

And, it has strong coding capabilities.

I am already inventing new games:

A screenshot of the Grok 3 chat interface on a light gray background, showing a conversation with a user. The user’s prompt reads, 'please create an entirely new game that I can play with arrows,' and Grok 3’s response details a new game concept called 'Arrow Flux.' The game is described as a puzzle-strategy game where players use arrow keys to navigate and manipulate a dynamic grid of energy flows to unlock an exit. The interface includes Grok 3’s logo, navigation icons, and a profile picture of the user in the top-right corner.

xAI will continue developing/ refining this model, and even better performance is anticipated.

It is very strong in coding (my limited amount of tests seem to confirm its superiority), and soon I will use it in building (AI) apps.

If you would like to learn with me, I suggest you subscribe to my Premium. We still have the promotion period where you can get additional no-cost access to my online course, Everyone can Code! ✨

I share my learnings with you, and keep you updated on AI/ how to build best with AI, as new tools/ AI will be launched.

I am launching the updated online course: Everyone can Code!✨ (Updated, because it keeps valuable lessons on working with AI, but has a chapter that I update constantly.)

🚨Special DEAL: Get a yearly Premium Subscription to this newsletter and receive the course Everyone can Code!✨for free!🚨 (Together worth 200 €)

(You will be emailed with the course entry. Current Premium subscribers will get access to the course automatically.)

Here’s Why Over 4 Million Professionals Read Morning Brew

  • Business news explained in plain English

  • Straight facts, zero fluff, & plenty of puns

  • 100% free

(✨ If you don’t want ads like these, Premium is the solution. It is like you are buying me a Starbucks Iced Honey Apple Almondmilk Flat White a month.)

On AI Evolution

A professional headshot of Martin Musiol, a man with short brown hair, a beard, and glasses, wearing a light blue button-up shirt. The image is framed with a pink and teal border, featuring the title 'A Conversation on the Evolution of GenAI: Insights from Martin Musiol' in large white text. Below the title, a bio describes Martin’s background in computer engineering, his experience at leading companies like IBM and Mphasis, and his role as the creator of the world’s first online course in generative AI, published by Wiley. The image is credited to DATA & AI MAGAZINE.

I hope you enjoyed it.

Happy weekend!
Martin 🙇

Our webpage

You might like our last episodes: