Grok 3: Better Than o3 and R1?

Which reasoning model should creators and entrepreneurs choose?

So we finally have the smartest AI on Earth. At least that's how Elon Musk describes the latest xAI model, Grok 3. Is that really the case? And does it mean it's time to cancel your ChatGPT subscription? Today we answer these questions.

In this issue:

• Overview: Grok 3 & Its Features

• Technical Comparison with o3-mini & DeepSeek R1

• Test Drive of Three Models

Keep your mailbox updated with practical knowledge & key news from the AI industry!

“Smartest AI on Earth” Is Revealed

As I mentioned above, before the release of Grok 3 (and even more so after) Musk did not skimp on ambitious statements. According to xAI, the new model is 10 times more powerful than its predecessor, leads in all parameters in academic tests and produces responses at an exceptional level. But loud words aside, we are dealing with a truly impressive product.

Here's why.

Grok 3 was trained on the XAI Colossus supercomputer, which includes about 200,000 GPUs. This amount of power allowed xAI to catch up and run the model with all the modern features, including “Thinking” (analog for ChatGPT’s reasoning), “Big Brain” mode, and DeepSearch.

• Thinking & Big Brain

"Think" Mode: Displays the chatbot's step-by-step reasoning process, enhancing transparency in responses.

"Big Brain" Mode: Allocates additional computational resources for complex tasks. It provides more detailed and accurate answers.

• DeepSearch

Grok-3 includes a built-in search engine called DeepSearch, enabling real-time information retrieval and the ability to articulate its thought process when responding to user queries.

xAI’s calls DeepSearch its first agent.

Grok 3 can also still generate images based on prompts, utilizing the Aurora model. Judging from my tests and what I've seen on X, the pictures have gotten more realistic.

Political and Cultural Aspect

The political and cultural side of the issue are worth mentioning separately. For Musk, these are fundamental aspects. According to him, Grok 3 has minimal censorship restrictions and can speak out on any topic. That said, xAI has trained it to make the model “based” as possible. Here's Grok’s definition.

You can see examples of reasoning on hot topics in the replays under this post.

Share this post with friends, especially those interested in AI Insights!

Availability and Price

Grok 3 is available through multiple tiers with varying pricing and access levels. As of February 20, free access to basic Grok 3 features is temporarily available to all users through X's platform and standalone apps, though with strict usage limits.

Free tier: 10 prompts & 10 images every 2 hours, three image analyses per day.

X Premium ($8/mo): Basic access to Grok 3, suitable for general use.

X Premium+ ($40/mo): Advanced features (Think, Big Brain, and DeepSearch) with higher usage limits.

SuperGrok iOS App ($30/mo): Same as for X Premium+ subscription.

Android app pre-registration is open, with full release imminent.

---

Grok 3 vs OpenAI’s o3 vs DeepSeek R1

Let's start with the technical part. Traditionally, for the AI industry, with the release of each new model, we get a series of screenshots from benchmarks, where the company boasts about its achievements. Grok 3 is no exception.

xAI tested two models, a basic and a mini, on the AIME 2025 and 2025 math exams. Both platforms outperformed all competitors, including OpenAI's o3 and DeepSeek R1. It’s similar to other tests, including GPQA and LiveCodeBench.

In the MMMU test, Grok scored two tenths of a point lower than o1.

You can compare the performance of each model in the image below.

However, I believe that while such images are great for press releases, things usually look different in real-world tasks. So now we're going to put the top three models head-to-head.

We'll compare them in three categories:

• Logic & Analysis

• Coding & Development

• Creative writing

After that, we'll summarize the results and determine the winner.

Logic & Analysis