🚀 Grok 4 Goes #1

PLUS: Google’s AI Cancer Drugs Near Human Trials

Author

The Summary AI
July 10, 2025

Welcome back!

xAI has released Grok 4 and Grok 4 Heavy, trained on its “Colossus” supercomputer, topping nearly every benchmark in reasoning, math, and science, and leapfrogging OpenAI, Google, and Anthropic in the ARC-AGI-2 test. The release arrives amidst controversy triggered by leadership exits and Grok 3 moderation incidents. Let’s unpack…

Today’s Summary:

  • 🚀 Grok 4 dominates frontier benchmarks

  • 🔍 Perplexity launches Comet AI browser

  • đź’Š Google AI cancer drugs to enter human trials

  • 📆 OpenAI to release AI web browser

  • 📉 YouTube preparing to crack down mass AI content

  • 🧑‍💻 ChatGPT tests “Study Together” mode

  • 🛠️ 2 new tools

TOP STORY

Grok 4 launches with top scores

The Summary: xAI launched Grok 4 and Grok 4 Heavy, claiming the top spot across nearly every major AI benchmark. The models beat OpenAI, Google, and Anthropic on the toughest tests in math, science, and reasoning. This launch could pressure rivals like OpenAI and Google to speed up their release timelines, including GPT-5.

Key details:

  • Scored 16.2% on ARC-AGI-2, nearly double that of Claude Opus 4

  • Scored 25.4% on Humanity’s Last Exam and 44% with tools

  • Grok 4 Heavy uses multiple agents to cross-check results

  • Handles 256k-token context

  • Voice mode delivers full answers naturally

  • Output speed is 75 tokens/second, faster than Claude Opus but slower than o3 and Gemini Flash

  • Available via API at $3/M input, $15/M output, with cache discounts

  • Consumer pricing starts at $30/month for SuperGrok and $300/month for SuperGrok Heavy (no free tier)

  • The release comes amidst controversy over Grok 3 moderation issues

Why it matters: Grok 4 is now a state-of-the-art model, proving how quickly the AI industry is moving. Just months ago, xAI was barely in the race, and now it’s outscoring the biggest players. This leap in performance validates the model’s reasoning techniques and multi-agent design. Pressure is rising on OpenAI, Google, and Anthropic to accelerate their next-gen releases. We could see GPT-5 or Gemini 3 land sooner than expected.

PERPLEXITY

Perplexity launches Comet Browser with built-in AI agent

The Summary: Perplexity has launched Comet, a new web browser with an integrated AI assistant that transforms online browsing into a seamless workflow. Instead of tabs and searches, users can ask questions, make purchases, send emails, or manage tasks directly from any page. With context management baked in, Comet aims to redefine the browsing experience.

Key details:

  • Comet is available today for Perplexity Max users ($200/month), with more waitlist invites rolling out through summer

  • The browser integrates Comet Assistant, an AI agent that can book meetings, compare products, summarize emails, and manage your browsing flow

  • During hands-on testing, the assistant could analyze YouTube videos, social media, and text inside Google Docs

Why it matters: Perplexity is aiming for the interface layer of the internet. Turning the browser into a task-executing agent transforms user behavior from clicking to instructing. Owning the browser also unlocks first-party data for the AI without relying on Google infrastructure. Comet, along with rumors of an incoming browser from OpenAI, is showing a return to browser innovation, and a real test of whether users will trust AI agents to manage their browsing.

FROM OUR PARTNERS

What Top Marketers Are Doing With AI

The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.

This guide distills 10 AI strategies from industry leaders that are transforming marketing.

  • Learn how HubSpot's engineering team achieved 15-20% productivity gains with AI

  • Learn how AI-driven emails achieved 94% higher conversion rates

  • Discover 7 ways to enhance your marketing strategy with AI.

GOOGLE

AI-designed cancer drugs near first human trial phase

The Summary: Alphabet’s Isomorphic Labs, born from Google DeepMind’s AlphaFold project, is preparing to test its first AI-designed drugs in humans. After a recent $600 million funding round and partnerships with pharma companies, the company says clinical trials are going to begin soon. Isomorphic’s goal is to create a drug design engine that combines machine learning with pharmaceutical expertise to streamline development.

Key details:

  • AI models can now simulate how a drug interacts with human proteins before it’s made, moving drug design from the lab bench to algorithms

  • One Isomorphic Labs prototype drug was designed entirely in silico until preclinical validation

  • Cancer and immunology drugs are the first in the company’s internal pipeline, nearing human trials

  • $600 million raised in April 2025 in its first external funding round

  • Collaborations signed with Eli Lilly and Novartis for co-developing the drugs

Why it matters: Pharma bets billions on drug discovery with just a 10% trial success rate. If Isomorphic AI can improve that rate, it will cut years and costs out of development. AlphaFold leap from static protein prediction to interaction modeling gives researchers a live simulation tool before human testing even begins.

TOOLS

🥇 New tools

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/