Good morning, AI enthusiasts. xAI’s highly anticipated Grok 4 has arrived — and it’s crushing benchmarks across the board.

The newest “truth-seeking” AI takes us a step closer to AGI, with Musk suggesting it may start discovering new physics “as soon as this year.” But given the backlash over Grok 3’s recent racist and antisemitic comments, the release is also likely to face more scrutiny than ever before.

In today’s AI rundown:

• xAI releases Grok 4 following 3’s crashout

• Perplexity’s Comet browser for AI-first web

• Turn messy image filenames into descriptive ones

• OpenAI snags top engineers from rivals for scaling team

• 4 new AI tools & 4 job opportunities

LATEST DEVELOPMENTS

XAI

🚀 xAI releases SOTA Grok 4 following 3’s crashout

Image source: xAI

The Rundown: xAI just announced Grok 4 and Grok 4 Heavy, its next-gen reasoning-only models that are “better than PHD levels in every subject” and deliver SOTA capabilities across benchmarks, including Arc-AGI and Humanity’s Last Exam.

The details:

• Grok 4 is a single-agent AI with voice, vision, and a 128K context window, while 4 Heavy is its advanced sibling, with multiple agents to tackle complex tasks.

• Both mark a major jump in benchmarks, achieving SOTA on Humanity's Last Exam, Arc-AGI-2, and AIME, and surpassing Gemini 2.5 Pro and OpenAI’s o3.

• Grok 4 is available with the SuperGrok subscription at $30/month, while Grok 4 Heavy is part of the new SuperGrok Heavy plan priced at $300/month.

• The new model is also available via API with a 256K-token context window and built-in search, priced at $3/million input tokens and $15/million output tokens.

• The power-packed release comes after a major backlash against Grok 3, which was caught making racist and antisemitic comments after an update.

Why it matters: Despite being a relatively new player, Musk’s xAI is already challenging the AI heavyweights. The latest release showcases the power of its Colossus supercomputer and pushes the scaling frontier further, though in the wake of the Grok 3 controversy, it’s likely to face heightened scrutiny from experts around the world.

TOGETHER WITH ASAPP

♻️ Automate complex customer interactions with AI agents

The Rundown: ASAPP’s GenerativeAgent is an enterprise-grade AI agent that resolves real customer issues on its own across voice and chat. Built for high-stakes environments, it goes beyond conversation to deliver real outcomes backed by security, collaboration, and continuous learning.

With GenerativeAgent, you’ll experience:

• Fewer transfers, faster resolutions with human-AI collaboration workflow

• Built-in guardrails and enterprise-grade data protection

• Real-time quality monitoring and self-improving

Take a self-guided tour of GenerativeAgent today.

PERPLEXITY

🖥️ Perplexity’s Comet browser for AI-first web

Image Source: Perplexity

The Rundown: Perplexity launched Comet, a new AI browser that embeds the company’s search engine alongside an assistant capable of performing agentic tasks—like booking meetings and navigating websites—while integrating with user workflows.

The details:

• The Comet Assistant lives in a sidebar that watches users browse, answering questions while automating tasks like email and calendar management.

• Users can utilize the agentic assistant to “vibe browse” without interacting directly with sites, using natural language or via voice commands.

• The browser promises seamless integration with existing extensions and bookmarks, supporting both Mac and Windows at launch.

• Perplexity Max users ($200/mo subscription) get first access along with a rolling waitlist, with Pro, free, and Enterprise users coming at a later date.

Why it matters: Chrome has had a chokehold on the browser for years — but appears to be a step behind on the agentic, AI-driven transition. While there will be hiccups as agents continue to evolve, Dia, Comet, and soon OpenAI (more below) are taking the first steps into a new, inevitable shift in how we navigate and take actions on the web.

AI TRAINING

📸 Turn messy image filenames into descriptive ones

The Rundown: In this tutorial, you will learn how to use Google's Gemini CLI to analyze your images and generate SEO-friendly filenames automatically, improving your content organization and search engine visibility.

Step-by-step:

1. Install Gemini CLI: npm install -g @google/gemini-cli and authenticate with your Google account

2. Test a single image analysis typing: gemini "Describe what's in [image1.png]"

3. Batch process: gemini “Process all images in this folder. For each image, analyze the content and rename it with a descriptive filename with relevant keywords for SEO purposes.”

Pro tip: Start with a small batch to understand how Gemini interprets your content, then scale up to your entire image library.

PRESENTED BY INVISIBLE

📊 AI training trends from those behind the scenes

The Rundown: As the partner for many enterprises and leading foundation models including AWS, Microsoft, and Cohere, Invisible has a front row seat into how the AI training landscape is evolving in real-time.

In this free guide, Invisible reveals:

• The most important shifts happening in model development, optimization, and deployment

• The rising bar for quality data

• A focus on specialized training sets

• The emergence of agentic products for real-world outcomes

Download the free guide.

OPENAI

🥊 OpenAI snags top engineers from rivals for scaling team

Image source: Greg Brockman (@gdb on x)

The Rundown: OpenAI recruited four new senior engineers from Tesla, xAI, and Meta for its scaling team, according to WIRED — joining to work on the Stargate data center and infrastructure initiatives, and coming during a tense AI talent war with tech giants.

The details:

• Former Tesla VP of software engineering David Lau will oversee OAI’s backend systems, revealed in an internal message from co-founder Greg Brockman.

• Engineers Uday Ruddarraju and Mike Dalton join OAI’s scaling team to work on Stargate after helping build the 200,000-GPU Colossus supercomputer at xAI.

• Former Meta AI researcher Angela Fan also joins the scaling team, coming amid Meta’s aggressive recruitment of OAI staff that has poached seven staffers.

Why it matters: These hires mark the first public moves for OpenAI since Meta’s hiring spree that has poached talent from across the AI leaders. It’s also a direct strike at Elon Musk’s engineering crew from xAI and Tesla — and given the past relationship between the two, it may stir the pot even further in their ongoing feud.

QUICK HITS

🛠️ Trending AI Tools

• 🎬 Marey - Filmmaker-focused AI video model trained on licensed content

• 📽️ Veo 3 - Google’s video AI, now with first-frame image + audio output

• 🪪 Higgsfield Soul ID - Personalized, consistent character image generation

• 🤝 Coachvox - Clone an AI version of yourself to coach clients in your style

💼 AI Job Opportunities

• 🧠 Databricks - Director, Emerging Enterprise

• 📈 Harvey - Senior Analytics Engineer

• 🗂️ Notable - Technical Program Manager

• 🔬 Deepmind - Data Scientist, GeminiApp, Verticals

📰 Everything else in AI today

Get up to speed on Agentic AI – learn how to build, test, and deploy AI Agents with Postman’s Rodric Rabbah in this free, on-demand webinar.*

OpenAI is set to launch its own web browser in the “coming weeks” that will challenge Google Chrome, featuring a ChatGPT-like chat interface and agentic integrations.

OpenAI will also reportedly release its highly anticipated open-source model next week, rumored to be “similar to o3 mini” with reasoning capabilities.

Microsoft CCO Judson Althoff said the company has saved over $500M in the past year from AI’s infusion in call centers, following last week’s cut of 9,000 jobs.

AI2 introduced FlexOlmo, a new language model training paradigm that enables data owners to contribute to AI development without sharing their raw data.

Google integrated Gemini into WearOS smartwatches from Pixel, Samsung, Xiaomi and more, enabling natural voice interactions and task management on the devices.

OpenAI announced that its acquisition of Jony Ive’s firm, io, has closed, with Ive and his LoveFrom team staying independent but embedded in OpenAI’s design direction.

*Sponsored Listing

COMMUNITY

🎥 Join our next live workshop

Join our next workshop this Friday, July 11th, at 4 PM EST with Dr. Alvaro Cintas, The Rundown’s AI professor. By the end of the workshop, you’ll confidently be able to install and use Gemini CLI to boost your productivity right from the command line.

RSVP here. Not a member? Join The Rundown University on a 14-day free trial.

See you soon,

Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team