With the flurry of AI news penetrating the headlines in recent weeks, you might’ve missed Elon Musk’s intent to launch Grok 3, xAI’s most advanced large language model (LLM) yet. The big day has finally come, as Musk announced the release of Grok 3 in beta to all Premium Plus members today.
If you weren’t aware, that’s a whole tier above the Premium subscription, which still raids your bank account every month in exchange for fewer ads and less restrictive access to Grok 2 – xAI’s more widely available generative AI model. Grok 3 isn’t quite as generous, unfortunately.
Taking Grok to a new level
— Elon Musk (@elonmusk) February 18, 2025
While we haven’t tested Grok 3’s capabilities, not being a Premium Plus subscriber, Musk has made some rather bold claims about the performance of his new model. In January, Musk claimed that the incoming Grok 3 was kitted out with at least ten times more computing power than its predecessor.
In terms of raw ability when it comes to maths, science, and coding, Grok 3 beats them all – including Google’s Gemini, ChatGPT’s 4o model, DeepSeek’s V3, and Anthropic’s Claude according to Musk’s own benchmarks.
But then of course he would make some outlandish claims that can’t immediately be debunked by his doubters, limiting access to those willing to pay a monthly subscription to their favourite social media website. Still, having tested Grok 2 with some success, we’re keen to get our hands on the new bot to see what it’s really made of.
Read More: Google Gemini just got a whole lot smarter, knows what you did last summer
When that’ll be, however, remains a mystery. We reckon it’ll be a few months at the very least before Musk is willing to give up Grok 3 access to his, cough, Premium-or-lower users. Musk clearly intends to ride the hype train for some time, considering the just-increased price of Premium Plus in the U.S., following another hike in December.
If you are one of Musk’s loyal Premium Plus subscribers, you’ll have a choice of three Grok 3 models. There’s the standard version that’ll compete with the likes of ChatGPT-4o and Gemini Advanced, coupled with Grok 3 mini that’ll stick with the simpler questions and faster answers. Grok 3 Reasoning is the final model, meant to be a proper competitor to DeepSeek’s highly impressive DeepSeek R1 model.
According to Musk’s launch live stream, the plan is to open-source all previous Grok models once Grok 3 is “mature” and ready. He didn’t offer any specific timeframe, mentioning that he expects that process to take a couple of months at least.
