Musk’s xAI releases Grok 3 models, claims superiority against competitors

The Grok 3 family of models will power Deep Search, allowing users to do multi-step research.

Elon Musk’s company, xAI, has unveiled its Grok 3 family of models and started rolling out its new features for Premium subscribers on X, formerly known as Twitter.

Musk and a few researchers announced via live stream video on X on Monday that the latest Grok 3 model used more than ten times the computing resources of Grok 2.

xAI claims that Grok 3 surpasses competitors such as OpenAI, Anthropic, and DeepSeek in several categories, such as general mathematical reasoning, knowledge about STEM and science, and coding.

The early version of the model, called Chocolate, is also the best overall model on lmarena benchmarks, where users evaluate models through computer-generated battles.

“Chocolate” was the first to score 1400 points on the website, surpassing Google’s Gemini family of experimental models, announced two weeks ago.

BREAKING: @xAI early version of Grok-3 (codename undefinedchocolateundefined) is now #1 in Arena! 🏆

Grok-3 is:
- First-ever model to break 1400 score!
- #1 across all categories, a milestone that keeps getting harder to achieve

Huge congratulations to @xAI on this milestone! View thread 🧵… https://t.co/p8z8lccNd5 pic.twitter.com/hShGy8ZN1o

undefined lmarena.ai (formerly lmsys.org) (@lmarena_ai) February 18, 2025

xAi also unveiled an agent Deep Search feature, allowing users to do more complex tasks such as research, saving “hundreds of hours of Google time.”

During the live stream, an example was shown typing a prompt like “When will be the next starship launch?” The model generated a list of articles providing sources, and xAI representatives said it could also verify whether the information was correct.

Stay informed and get our latest stories on Google News

Follow us

Recently, Open AI released a similar Deep Research feature powered by its o-3 model. This feature uses reasoning to synthesize information and complete multi-step research tasks.

Musk said that within a week, Grok 3, which started rolling out to users on the web, Android, and iOS, would include voice interactions.

According to Musk, when Grok 3 is fully out in a few months, the previous version will be made open-source.

Last week, a consortium led by Musk offered $97.4 billion to buy the nonprofit that controls OpenAI. Sam Altman declined, making a cheeky counter-offer to buy Twitter (now X) for $9.74 billion.