Just hours ago, yet another deep-thinking large language model hit the timeline, crushing existing benchmarks and reaching the coveted number one spot on the LM Marina leaderboard. This new model is none other than Elon's based and unfiltered Grok 3. Not only is it incredibly smart, but it’s also mostly uncensored and can generate content that is outright banned in many parts of the world.
What Makes Grok 3 Special?
Deep Thinking Mode & More
Grok 3 comes equipped with a Deep Thinking Mode, similar to DeepSeek, allowing it to process complex queries more efficiently. It can generate text-to-video, and XAI has teased an upcoming paid subscription for something even more powerful—Super Grok.
"Grok 3 is optimized for maximum truth-seeking, even if that comes at the cost of political correctness." – XAI Team
The AI Wars Heat Up
Last week, Elon Musk attempted to buy out OpenAI, trolling them with an offer that was promptly rejected. Meanwhile, Mark Zuckerberg took a major hit when it was revealed that Meta’s LLaMA models were trained on 82 terabytes of pirated books from the infamous Library Genesis Project.
"I just can't believe Zuckerbucks would do something like that... said nobody ever."
Exclusive Access to X Data
One major advantage Grok 3 has over competitors like GPT-4, Gemini, and Claude is its direct access to Twitter’s full firehose of data. Unlike its rivals, which rely on curated datasets, Grok is fed a continuous stream of real-time internet discourse, making it uniquely suited for generating up-to-date and culturally relevant responses.
Benchmarking Grok 3
Currently, Grok 3 sits at the top of LM Marina, a blind comparison leaderboard where human testers rank models side by side. It has also outperformed other state-of-the-art models in math, science, and coding tasks. However, some conveniently omitted benchmarks—like Codeforces and ARC AGI—raise questions about cherry-picking.
Head-to-Head Performance
Despite its strengths, all top-tier LLMs seem to be plateauing, shifting the AI race toward better prompting frameworks rather than just building larger models.
The World's Largest AI Supercomputer
Grok 3 was trained at XAI’s Colossus Supercomputer in Memphis, Tennessee—believed to be the largest AI cluster on Earth. It currently runs on 200,000 NVIDIA H100 GPUs, with plans to scale up to 1 million GPUs. Due to its massive power requirements, the facility had to bring in portable diesel generators to meet its energy demands.
The Future: Super Grok
XAI has hinted at an upcoming Super Grok subscription, expected to cost 200 per month. With a pricing war looming, AI companies are looking for ways to lock in users while improving their models.
Conclusion
Grok 3 is a major step forward, but is it truly the best? While it dominates some benchmarks and offers unparalleled access to real-time data, it remains to be seen if it can surpass OpenAI and DeepMind in the long run.
What do you think about Grok 3? Will it change the AI landscape forever, or is it just another step in the race? Let me know