Sadiq S BhatWashington, DC, — It began, as these things often do in Musk-world, just past midnight.
On a livestream beamed from a shadowed couch somewhere in California, Elon Musk unveiled Grok 4, his latest brainchild from xAI.
The world's richest man leaned in, surrounded by a trio of engineers, and declared it "better than PhD level" across every subject. "No exceptions."
The demo that followed had all the usual theatre: Grok 4 solving advanced math problems, visualising two black holes colliding, even predicting next year’s World Series winner.
xAI is branding it the "smartest AI in the world."
"Grok 4 is smarter than almost all graduate students in all disciplines, simultaneously," Musk said during the livestream. "That's really something."
Musk and his xAI crew say Grok 4 has outgunned the giants: beating Google's Gemini 2.5 Pro and OpenAI's o3 (high) on a gauntlet of tests they call Humanity’s Last Exam, a sprawling set of 2,500 crowdsourced questions spanning maths, humanities, and natural sciences.
Gemini still officially leads the leaderboard, but xAI claims Grok's latest leap pushes it past the pack.
A more powerful "Heavy" version of Grok, available for $300 a month, lets multiple AI agents collaborate on problems.
There's a new voice mode, too, designed to sound more natural. It struggled when prompted to deliver an opera about Diet Coke.
No review or transparency report
But for all the bombast, there was no mention of what had happened the day before: the previous version of Grok had gone off the rails.
Grok 3 briefly went rogue, posting remarks ranging from deeply insensitive to factually absurd and antisemitic.
The posts were subsequently taken down by X. While xAI issued a statement, Musk, in typical fashion, deflected. The bot was "too eager to please," he wrote, and would be "fixed."
Then, less than 24 hours later, Grok 4 arrived — framed not as a reset, but as a leap forward.
In the rush to outpace rivals like OpenAI and Anthropic, Musk seems to be betting on frequency over stability.
A coding-focused model is promised for August, a multimodal agent in September, and a video-generating model in October.
It's an AI arms race, and Musk has his foot pinned to the accelerator.
However, the cost of that speed is becoming increasingly difficult to ignore.
There was no technical report accompanying Grok 4's release. No published benchmark data. No independent review.
Where companies like OpenAI and Google offer detailed transparency reports, outlining what their models can do, and crucially, what they can't, xAI prefers a livestream and a few slide decks. Even those were short on specifics.
While Grok's meltdown this week wasn't entirely surprising to those watching closely, xAI's response, essentially to increase pre-post moderation, raises more questions than it answers.
Who decides what gets through? How will oversight evolve as the model becomes more complex and more autonomous?
And how exactly does "truth-seeking" work when deployed on a platform that regularly amplifies conspiracy theories?
Strategy and instinct
At $30 a month for standard access, Grok 4 is priced to compete with other consumer-facing AI tools. But xAI's ambitions stretch far beyond monthly subscriptions.
The company is pitching its technology to businesses, governments, and advertisers. Public filings reveal billions raised through venture capital and debt. A new data centre — Colossus — is under construction in Tennessee.
In that context, Grok isn't just a chatbot. It's a flagship product in an unfolding commercial strategy. But whether it's stable enough or trusted enough to serve serious institutional clients is far from clear.
Musk's messaging doesn't help. During the livestream, he veered from calling current AI tools "primitive" to claiming Grok could help discover new physics "later this year."
There was talk of usefulness, common sense, and honour — values, Musk said, "you’d want to instill in a child." The metaphor sat awkwardly next to Tuesday’s outrageous outputs.
It's also worth noting what Musk didn't say. Linda Yaccarino, CEO of X, quietly announced her departure just hours before the Grok 4 launch. No reason was given.
The release Grok 4 model is more than just a new spectacle.
What's being optimised here isn't just AI performance. It's attention and a loop that rewards disruption.
For Musk, that loop is both strategy and instinct.
Whether Grok 4 is "smarter" than its peers remains to be seen. Whether it's going to be safer, more stable, or more useful — those questions have not been addressed.
Musk, ever fond of grand analogies, framed Grok 4 in characteristically dramatic terms: "You can think of AI as this super genius child that ultimately will outsmart you."
