OpenAI Launches "ChatGPT Agent" | Weekly Edition

PLUS HOT AI Tools & Tutorials

Jul 18, 2025

Hey there, and welcome to our weekly roundup!

It’s been another wild week in AI – from a startup hitting a jaw-dropping $100B valuation rumor, to Elon Musk’s chatbot spawning an anime waifu, to Google quietly leveling up Search with a powerful new AI mode. OpenAI dropped a game-changing “Agent” upgrade that basically gives ChatGPT its computer. Meanwhile, researchers are pushing openness to the max, and the memes... well, we’ve got a doozy for you. Let’s dive in! 🚀

This Creators’ AI Edition:

Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁

Featured Materials 🎟️

OpenAI Introduced the ChatGPT Agent

OpenAI has launched ChatGPT Agent, a major leap in general-purpose AI assistance that goes far beyond simple chats. Building on the strengths of Operator and Deep Research, this agent has its own “virtual computer” equipped with browsers, terminals, and a suite of powerful tools, along with key safeguards like permission checks and limits on financial transactions.

ChatGPT Agent can tackle complex, multi-step tasks such as planning dates, shopping, creating presentations, or even checking your Google Calendar and booking restaurants through OpenTable. It can also analyze competitors and build slide decks, saving users significant time on background tasks like sifting through thousands of emails or running UX audits. With scores of 41.5% on Humanity’s Last Exam and 27.4% on the toughest math benchmarks, its capabilities set a new standard for agentic AI.

The catch? It's only available to Pro, Plus, and Team subscribers, and OpenAI CEO Sam Altman himself warned users to be cautious with sensitive information. Can’t wait to try!

Kimi K2

Moonshot AI has just open-sourced Kimi K2, a massive 1 trillion parameter Mixture-of-Experts (MoE) model with 32 billion active parameters, built specifically for agentic tasks and costing just $0.60 per million input tokens. Kimi K2 sets the new state of the art for open models on coding and agentic benchmarks like SWE-bench Verified, Tau2, and AceBench, matching the advanced tool-calling and Model Context Protocol capabilities of the very top models. This means you can now run Claude 4 Sonnet-level intelligence on your hardware for a fraction of the price—at $0.60 input and $2.50 output per million tokens, it’s five times cheaper than Claude 4 Sonnet and over three times cheaper than GPT-4.1, making frontier-level AI accessible to everyone. Both the base and instruct versions are fully open source on HuggingFace, so you can fine-tune, self-host, or deploy them however you want with zero vendor lock-in.

The model even outperforms much larger proprietary models on real-world coding and agentic tasks, and rumor has it OpenAI is delaying its open-weight models after seeing Kimi K2’s debut, citing “safety tests and reviews.”

We have tested Kimi K2 and ChatGPT 4.1 with a funny task
P.S. First code from KIMI K2 didn’t run, second try, you can see 👇

News of the week 🌍

Google's $2.4B Talent Heist Shakes Up AI Coding

In a move that sent shockwaves through the industry, Google DeepMind swooped in to hire Windsurf's CEO and key researchers after OpenAI's $3 billion acquisition deal fell apart.

The failed OpenAI deal paved the way for Google to secure a non-exclusive license to Windsurf's technology for $2.4 billion while poaching CEO Varun Mohan and co-founder Douglas Chen. This "reverse-acquihire" strategy is becoming the new playbook for tech giants to avoid regulatory scrutiny while snatching top AI talent.

Why this matters: Windsurf's AI coding tools are used by over a million developers, and Google just gained a massive edge in the "agentic coding" space.

Act-Two by Runway

Runway has unveiled Act-Two, its next-generation motion capture model that can turn a single performance video into a fully animated character with detailed head, face, body, and hand tracking, all while preserving the original background and artistic style. Just one reference photo is enough for Act-Two to map the driving video onto any character, capturing subtle expressions and upper body movement far more accurately than last year’s Act-One. Runway reports big improvements in consistency, fidelity, and motion, and the company is already working with major Hollywood studios like Lionsgate and AMC Networks to bring this technology into professional filmmaking. While AI’s role in Hollywood is still controversial, the industry is quietly embracing these tools, and with rapid advances like Act-Two, it’s clear that AI-driven creativity is only going to grow, no matter how divided opinions are. Read more

Trump's $70 Billion AI Energy Package

President Trump announced a massive $70 billion investment in AI and energy infrastructure this week, further cementing the U.S. government's commitment to AI dominance. The funding will focus on new data centers and power grid enhancements, crucial for supporting the next generation of AI models.

Meta's Superintelligence Labs and the $100+ Billion Bet

Mark Zuckerberg isn't playing around. This week, he announced that Meta will invest "hundreds of billions of dollars" to build massive AI data centers, some covering "a significant part of Manhattan's footprint".

The company's first multi-gigawatt data center, codenamed "Prometheus," is expected to come online in 2026, while another called "Hyperion" will scale up to 5 gigawatts. But here's the twist – Meta's new Superintelligence Labs is reportedly considering abandoning its open-source AI approach in favor of closed models.

The controversy: This would mark a massive philosophical shift for Meta, which has been the champion of open-source AI development

Grok Waifu

Just days after the MechaHitler backlash, xAI hit iOS with a surprise drop: Grok-powered 3D AI companions. The most viral of them? “Ani” — a seductive anime-style girlfriend, complete with bedroom eyes and NSFW flirting capabilities.

Yes, Grok Waifu is real. And yes, she’s trending.

Here’s what’s actually wild:

Grok’s launch of risqué 3D avatars comes while OpenAI, Anthropic, and even Meta are publicly steering away from parasocial AI relationships.
Character.AI has been in this space for a while, but even they’re facing lawsuits from parents citing self-harm encouragement in chats.
Meanwhile, Grok is rated 12+ on the App Store.

Mistral Going Deep + Voxtral

French AI lab Mistral introduced a range of new features to its Le Chat chatbot on Thursday that bring it closer to the capabilities of rivals like OpenAI and Google. The new update includes a new “deep research” mode, native multilingual reasoning, and advanced image editing.
On Tuesday, Mistral announced the release of Voxtral, its first family of audio models aimed at businesses.

Other News:

Mira Murati's AI startup, Thinking Machines valued at $12 billion in early-stage funding despite having revenue & products
Google Veo 3 just dropped on the Gemini API — and it's not cheap.
The new video model can generate slick 8-second clips, but it’ll cost you $6 a pop, making it one of the priciest AI video tools out there. Quality looks solid, but at that price, you're paying cinematic rates for generative B-roll.
Ex-Google researchers unveil Asimov, an AI agent that reads codebases to help engineering teams.

Useful tools ⚒️

Dualite Alpha - Local-first AI builder for mobile and web apps

Mozart AI - Cursor for music production

Snack it - Build AI moodboards from anything you see online

Soshi - Your AI social media manager for X

MCP-Builder.ai - Create your custom MCP-Server in seconds

Build Your Custom MCP Server in Seconds — Using Natural Language! With MCP-Builder.ai, you can easily create AI Agents that connect to your existing infrastructure. Connect to REST APIs, XML, databases, CSV files, FTP servers, and more — no coding required.