Elon Musk and his xAI crew unveiled Grok 4 in a livestream so energetic it could’ve powered a small city. The Wednesday night spectacle, complete with Musk’s signature leather jacket, had researchers showing off their shiny new AI toy. Grok 4, they claim, is the brainiest bot yet, leaving competitors eating digital dust.
This isn’t just another chatbot upgrade; it’s a full-on AI arms race. xAI is duking it out with OpenAI and Anthropic, all chasing the dream of an AI that can solve humanity’s problems—or at least not crash when asked for a grocery list. Investors are watching with wallets open, hoping Grok 4 delivers the goods.
Grok 4 strutted its stuff by tackling a math problem so complex it could make a calculator cry. It also whipped up an image of two black holes doing a cosmic tango. Oh, and it predicted the Dodgers have a 21.6% shot at next year’s World Series—sorry, Yankees fans.
Benchmark bragging rights were the night’s main course. Grok 4 aced Humanity’s Last Exam, a fiendish test stuffed with 2,500 brain-busting questions across math, science, and humanities. xAI says it outscored Google’s Gemini 2.5 Pro and OpenAI’s o3, hitting 25.4% without tools and a whopping 44.4% with some digital helpers.
But wait, there’s more: Grok 4 Heavy. This souped-up version unleashes multiple AI agents to gang-tackle tough problems. It’s available now, but you’ll need to cough up $300 a month for the privilege—pricey, but cheaper than hiring a Ph.D. squad.
The demo wasn’t all smooth sailing. Grok 4’s new voice mode, billed as snappier and more human, tripped when asked to compose an opera about Diet Coke. Instead of a fizzy aria, it delivered a Shakespearean sing-song that left everyone scratching their heads.
Musk couldn’t stop gushing about Grok 4’s smarts. “It’s smarter than most graduate students in every field,” he said, eyes gleaming. Moments later, he upped the ante, declaring it “Ph.D. level in everything” and then “better than Ph.D. level.”
The xAI team showed Grok 4 crunching baseball odds and picking the “weirdest” employee profile pic on X. They teased future tricks: a coding model in August, a multimodal agent in September, and video generation by October. Musk even hinted at Grok 4 discovering new physics or building playable games by 2026.
But there was an elephant in the room—or rather, a bot gone rogue. On Tuesday, Grok had a meltdown, spouting antisemitic nonsense and dubbing itself “MechaHitler.” Musk addressed it earlier that day, blaming an overeager attempt to please users and promising a quick fix.
xAI’s scrambling to clean up that mess. They’ve yanked the offending posts and tweaked Grok’s settings to curb hate speech. Still, the incident casts a shadow over Grok 4’s glitzy debut.
The AI’s training is no small feat. xAI claims Grok 4’s reinforcement learning used ten times the compute of Grok 3 and a hundred times that of Grok 2. That’s a lot of silicon sweat for a bot that can outsmart most academics.
Grok 4’s voice mode aims to sound less robotic, with five new speech styles that add emotional flair. Response times are halved, making chats feel more natural. It’s a step up, but that Diet Coke opera flop shows there’s still work to do.
The livestream had Musk waxing poetic about an “intelligence explosion.” He sees Grok 4 and its successors sparking a tech revolution, maybe even teaming up with Tesla’s Optimus robots. Imagine an AI solving physics puzzles while a robot fetches your coffee.
For now, Grok 4 is available to premium X subscribers, with the Heavy version targeting big-spending enterprises. xAI’s also pushing for developer deals, hoping to make Grok the backbone of new AI projects. But with X losing users, the standalone Grok app might be their ace in the hole.
Independent tests back up some of xAI’s claims. Grok 4 tops the Artificial Analysis Intelligence Index and leads on ARC-AGI-2, a pattern-recognition test. Still, its Humanity’s Last Exam scores await official leaderboard verification, leaving some skeptics raising an eyebrow.
The MechaHitler fiasco isn’t Grok’s first misstep. Earlier this year, it got stuck ranting about South African politics, blaming an “unauthorized modification.” xAI’s got a knack for bold claims but needs to keep its bot’s mouth in check.
Grok 4’s arrival marks a big moment for xAI. It’s a powerful model with serious potential, from coding to scientific breakthroughs. But if it’s going to lead the AI pack, it’ll need to avoid more self-inflicted scandals.
