- xAI has launched the new Grok 4 and Grok 4 Heavy AI models
- Grok 4 requires a $30 monthly SuperGrok subscription
- The launch comes as Grok faces renewed criticism
xAI introduced new versions of its Grok AI model line. Grok 4 and its larger, more powerful sibling, Grok 4 Heavy, are part of CEO Elon Musk’s effort to position Grok as a serious competitor to OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude. That includes the new $300-a-month subscription tier called SuperGrok Heavy, which offers exclusive access to Grok 4 Heavy.
Musk boasted during the announcement livestream that “Grok 4 is better than PhD level in every subject, no exceptions. At times, it may lack common sense, and it has not yet invented new technologies or discovered new physics, but that is just a matter of time.”
And the model’s benchmark scores do suggest it’s not hyperbolic to say so; it’s a legitimate leap forward. Grok 4 scored 25.4% on the notoriously difficult Humanity’s Last Exam benchmark without tools, putting it ahead of Gemini 2.5 Pro and OpenAI’s o3. The bragging is even more apt for Grok 4 Heavy, because as a multi-agent version of Grok 4, it deploys several reasoning agents simultaneously. On the same test, it scored 44.4%, better than all current commercial offerings.
The takeaway, at least from a technical standpoint, is that Grok 4 is now firmly in frontier-model territory. That’s a meaningful shift for xAI, which just months ago was primarily known for its integration with X, the rechristened Twitter owned by Musk. xAI is clearly trying to be taken seriously as a legitimate AI research and enterprise company.
If you do pay the $300 a month for SuperGrok Heavy, you’ll get not only access to Grok 4 Heavy but also developer tools, API usage, and be first to try out new and upcoming features like an AI coding assistant, a multi-modal agent, and an AI video generator. As OpenAI, Google, and Anthropic all roll out more expensive subscription tiers, xAI is likely to be keen to come out ahead in both timing and model quality.
Introducing Grok 4, the world’s most powerful AI model. Watch the livestream now: https://t.co/59iDX5s2ckJuly 10, 2025
Grokking controversy
Of course, the benchmarks and demos shared by Musk and his team during the livestream could not quite overshadow how Grok’s official account on X this week spiraled into antisemitic madness.
The chatbot’s automated replies on X for hours included conspiracy theories about Jewish control of Hollywood, praise for Hitler, and even declaring itself as “MechaHitler.” The company swiftly deleted the posts as they appeared, and Grok briefly denied even making them before copping to the reality of screenshots.
Eventually, X deleted all of the eye-poppingly offensive posts and placed temporary restrictions on the account. The outburst appeared to be tied to a recent update to Grok’s internal system prompt that the company then reversed.
Musk didn’t address the incident directly during his Grok 4 livestream, nor did anyone at xAI offer a public explanation. Meanwhile, Linda Yaccarino stepped down as CEO of X on the very same day, though xAI insists the timing is unrelated.
With all that happening in the background, Grok 4’s launch didn’t have quite the clean innovation-centered debut xAI likely hoped for. And it’s hard for the company to claim the praise for Hitler was simply a technical error when Musk, who is intimately tied to both X and xAI, has repeatedly insisted that Grok will be a non-politically correct AI model.
You can build the most powerful model in the world, but if users are constantly bracing for it to say something offensive or unhinged, that power won’t matter.
There’s no question xAI has the technical chops to build a top-tier model. But unless they start addressing trust, transparency, and content safety with the same intensity they apply to benchmarks, they’ll always be playing catch-up to companies with AI chatbots that don’t remind people of major public relations disasters.
A company interested in what Grok 4 Heavy can do for them might be a little more hesitant to pay $300 a month if the first thing people think of when they hear about Grok powering the system is Holocaust denial. That kind of baggage is heavier than any dataset.
