Elon Musk's xAI Breaks Benchmarks With Grok 4

Welcome to Lore Brief, your weekly edge in the age of AI.

This issue is brought to you by Factory, an engineer in every tab.

Factory is a Sequoia-backed startup that has created what I think is the best agentic coding platform out there. It’s better than Devin in all the tests I’ve done.

Whether you're prototyping quickly or building enterprise-grade solutions, Factory accelerates development. Major companies like Zapier, Bayer, MongoDB, and Framer use Factory to develop faster and solve more bugs, giving engineers time to focus on what they enjoy doing, not debugging.

64 K-view demo: watch me and founder Matan Grinberg clone the core DocuSign functionality in 15 minutes!

Try Factory at Factory.ai and follow Factory on Twitter/X for more updates!

xAI’s Grok 4 is finally here with stellar benchmark results

xAI announced Grok 4 on July 9th with genuinely impressive benchmark results establishing it as the new state-of-the-art model, at least on academic measures. New SOTA models only emerge a few times per year, making this a notable milestone in AI development.

Here’s a thread we created showing the top examples of Grok 4 in action.

Outstanding performance: 100% on mathematical olympiad problems. AIME competitions are used to select teams for international math olympiads and challenge the brightest mathematical minds at the high school level. Perfect score means solving every problem correctly.

But the real standout is "Humanity's Last Exam" - designed to be unsolvable by humans. Grok 4 scored 44.4% while the second place was for Gemini 2.5 Pro with 26.9%. We're talking about questions so difficult PhD-level experts can't reliably answer them.

The graduate physics benchmark hit 88.9%, and ARC-AGI-2 - which tests intuitive reasoning easy for humans but nearly impossible for AI - saw nearly a 2x improvement over previous records.

Elon Musk's quote "With respect to academic questions, Grok 4 is better than PHD levels in every subject. No exceptions." isn't hyperbole when you look at these benchmarks.

What's remarkable is xAI's trajectory. They used roughly 100x more compute for Grok 4 compared to Grok 2, with a breakthrough allocation: equal compute for pre-training and reinforcement learning, versus the traditional <10% for RL. This represents a fundamental shift in training methodology.

Musk's massive infrastructure buildout - that 100,000 GPU facility - clearly played a crucial role in enabling this computational scale.

API pricing sits at $3/$15 per million tokens, while consumer access costs $30/month for SuperGrok and Grok 4. For Grok 4 Heavy - the multi-agent model - the pricing matches the name at $300/month.

Whether this translates to practical advantages in real-world applications remains to be seen, but on academic benchmarks, xAI just claimed the crown.

Elon is hyping it hard, which isn't unusual for him, claiming he expects the next versions of Grok to start making real scientific breakthroughs in 6 months to 1-2 years. We'll see if this materializes, but is it really hype if it turns out to be true?

Google’s Veo 3 new, impressive feature: Lip sync voices

The breakthrough: Veo 3 generates videos with characters that actually speak, complete with dialogue, ambient sounds, and realistic lip synchronization, all from a text prompt and a single image.

The model excels at understanding complex narrative prompts and can generate 4K resolution videos with realistic physics and natural motion. Reference-powered video capabilities allow users to maintain character consistency across scenes by providing photos of characters or style examples.

Access has expanded significantly since launch. Veo 3 is now available to Google AI Pro subscribers ($20/month) in over 150 countries, with a limit of three videos per day. Google AI Ultra subscribers ($250/month) get the highest access with unlimited usage and advanced features. Pixel 9 Pro users recently received a free one-year AI Pro subscription, providing access to Veo 3.

The broader landscape is heating up. Products like Midjourney's video capabilities and Kling 2.1 are delivering increasingly impressive results. 

What makes Veo 3 particularly significant is its native audio-video integration. Previous AI video generators required creators to handle audio separately - generating silent clips then adding sound effects, dialogue, and music in post-production. Veo 3 eliminates this workflow bottleneck by producing complete video content with synchronized dialogue, ambient sounds, and proper lip-syncing from a single prompt.

Zuck's Multi-Million AI Talent Vacuum Continues Relentlessly

Meta just paid $200 million to poach Apple's AI chief Ruoming Pang. That's more than Tim Cook's entire compensation package. And Pang isn't even the biggest fish - they've already netted 10+ OpenAI researchers, plus top minds from Anthropic and Google DeepMind.

This looks like calculated warfare. Zuck's systematically gutting every AI lab in the Valley.

The numbers: $100M signing bonuses are becoming standard. Meta's new Superintelligence Labs now houses former GitHub CEO Nat Friedman, Scale AI's Alexandr Wang, and a who's who of AI big brains.

What if Zuckerberg actually knows exactly what he's doing? What if consolidating this much talent under one roof - with unlimited compute and data - creates an unstoppable gravity well?

The implications: OpenAI, Anthropic, even Google should be sweating. When one player can outbid everyone by 10x, the game changes. Today it's Pang. Tomorrow it could be entire teams.

Bottom line: Zuck's anaconda-sized appetite for talent shows no signs of slowing. Whatever he's building, it's going to be massive.

How John Rush Is Building a Zero-Employee Business

In this week’s episode of The Next Wave, John Rush explains how he runs 80% of his zero-employee company using custom AI agents, from automated SEO content creation to product prototyping, and why he believes the future belongs to founders orchestrating fleets of specialized AI workers. → Watch | Listen

Things I’m Learning From

• [Link] – List of all the new AI browsers (there are 6 of them!).

• [Link] – A new top trending band on Spotify… that’s entirely AI-generated!

• [Link] – Marc Andreessen’s career advice for the AI era.

That’s it for today.

Consider forwarding Lore Brief to a colleague to help them get ahead in the AI Age.

-Nathan Lands
Connect → X | LinkedIn
Listen to The Next Wave → Apple | Spotify | YouTube

(Disclosure: I may own equity in companies mentioned in Lore Brief.)