Grok is Playing Big

Last week, Grok unveiled “the smartest AI model,” Grok 3, which topped nearly all categories in initial benchmarks. So, what do end users think about it?

Elon Musk recently shared on X that Grok 3, touted as “the smartest” model, would launch on February 17. After its release, it quickly claimed the top spot across almost all categories on the LMArena leaderboard, sparking widespread curiosity among users. Currently accessible via X.com and Grok.com, the model has been tested by users and has largely left them impressed. But how does it hold up in long-term use?

I’ve been using Grok 3 extensively since its release. Overall, I’m satisfied with its DeepSearch and Think features. However, when things get heavy, I’ve personally seen Grok 3 start to falter. After a while, it gives wildly inconsistent answers—reminiscent of the infamous “hallucinations” seen in GPT models. X users have already begun poking fun at Elon Musk over this, with one post racking up hundreds of likes: “Just like its owner, wildly unstable and constantly shifting.” It’s hard not to agree.

Despite the mockery, calling Grok 3 one of the top 3 most powerful models wouldn’t be an overstatement. From my personal tests, it’s easily the best AI for reasoning through technical topics. In coding, it delivers solid outputs, though it clearly lags behind Anthropic’s Claude 3.5 family.

Pricing-wise, it’s not exactly a thrill. As an X Premium user, I can access all of Grok 3’s features, but I keep seeing a banner on Grok.com pushing the SuperGrok package. It promises higher rate limits and access to Grok 3 Thinking and DeepSearch—things I already have. I don’t get what it’s for, especially at $30/month.

The Grok family wasn’t on my radar before Grok 3, but after this big leap, I’m eagerly awaiting future updates.