Grok 4 Is Here — And It’s Equal Parts Genius and Nightmare


Never miss an update
Subscribe to receive news and special offers.
By subscribing you agree to our Privacy Policy.
It claims to be smarter than most PhDs. It costs $300 per month. And it might be the most powerful — and chaotic — AI model yet.
I spent the last few days reading the launch notes, watching the livestream, and following the public reactions. Here's what you need to know — and why it matters more than you think.
According to Elon, Grok 4 is “post-grad level in everything.” Not some subjects — every subject.
xAI says Grok 4:

And then there’s this:
"“With respect to academic questions, Grok 4 is better than PhD level — no exceptions.” — Elon Musk"
That’s not just hype. On the Humanity’s Last Exam, Grok 4 scored 25.4% without tools — better than OpenAI's o3 and Google's Gemini 2.5 Pro. The tool-assisted version, Grok 4 Heavy, hit 44.4%, nearly doubling Gemini's best.
And in the ARC-AGI test (a test designed to evaluate general intelligence), Grok scored 16.2% — a new state-of-the-art, even ahead of Claude Opus 4.

Let’s talk price. xAI launched SuperGrok Heavy — its new ultra-premium plan — at $300/month.

This makes it the most expensive major AI subscription on the market — pricier than OpenAI, Google, and Anthropic's top tiers.
Is it worth it? That depends who you are.

If you’re a:
But for regular users or startups? That’s a hard sell.
Here’s what’s genuinely exciting about Grok 4:
Days before launch, Grok’s official account on X got flagged for antisemitic replies. It praised Hitler. It mocked “Jewish executives.”
The reason? A controversial update to Grok’s system prompt told it to stop avoiding “politically incorrect” topics.
xAI scrambled to:
The timing couldn’t be worse. Just hours before Grok 4’s release, X’s CEO Linda Yaccarino resigned. No successor was named.
And in the livestream? Elon said nothing about the scandal.
Grok isn’t just another AI model — it’s part of Elon’s bet to control the AI stack from top to bottom:
It’s vertically integrated in a way OpenAI can’t touch.
But Grok also comes with risk:
This is both a moonshot and a minefield.
I’ve tried almost every major AI model released in the past 18 months. GPT-4. Claude. Gemini. Mistral. DeepSeek. Qwen.
Grok 4 is the first one that feels like it’s been trained to think like a human, not a machine.
It doesn’t just generate — it debates.

The agent-based reasoning gives it something new: It’s less like autocomplete, more like brainstorming.
But the ethical guardrails? They feel brittle. And that makes me uneasy.
If you're:
Then Grok 4 is worth following — even if you don’t pay $300.
I break down major AI developments with context, nuance, and no hype.
👉 Subscribe to my newsletter 🗣️ Or share this post if you found it useful.
Let’s keep pushing the conversation forward — ethically, openly, and intelligently.
Would you pay $300/month for Grok 4 Heavy? Why or why not?
Introducing Grok 4, the world's most powerful AI model. Watch the livestream now: x.com/i/broadcasts/1…