🦙 Llama 4 is here

Welcome back, Superhuman. Have you herd the news? Okay, terrible pun — but worth it, because a powerful new family of Llama 4 models just dropped over the weekend. And: I have good news and bad news about GPT-5. To pull the band-aid off quick, it’s taking longer than expected, but there’s a surprising upside.

Today’s Insights

• Llama 4, Copilot reimagined, Midjourney’s V7, and GPT-5’s roadmap

• Everything you need to know about Llama 4

• Tutorial: How to use agents for competitive analysis

• 5 new AI tools to boost your productivity

• News, memes, what’s trending on socials, and more

TODAY IN AI

1. Meta just dropped Llama 4: The social media giant showed off three versions of its long-awaited Llama 4 model, which each offer “unrivaled speed and efficiency” while being open to researchers. More details below.

2. Microsoft celebrates 50 years with revamped Copilot: The company just jolted its Copilot chatbot back into relevancy with some major updates. It can now remember details about your life, including food preferences, hobbies, and friends’ birthdays. And an “actions” tool can complete tasks for you, like buying tickets or filling out forms. Here’s a video with even more features, including AI-generated podcasts, vision capabilities, deep research, and customization.

3. Midjourney heats up image-gen race with V7: Almost a year since its last major upgrade, Midjourney is back with V7 Alpha, which was built on an entirely new architecture and, for the first time, lets you generate images just by talking to it. Plus, a new Draft Mode generates images at 10x the speed so you can brainstorm before landing on a final design. To try it out, you’ll first need to create a personalization profile here.

4. OpenAI is scrapping its original GPT-5 roadmap: CEO Sam Altman revealed that GPT-5 is still months away because it was harder than anticipated to integrate all of the company’s models into a single product. On the plus side, GPT-5 will be “much better” than even OpenAI was expecting. And the cherry on top: The full version of o3 as well as o4-mini are going to launch as separate models after all, and they’re only a couple of weeks away.

PRESENTED BY SPINACH AI

Still doing manual tasks in 2025? There’s AI Agents for that

Spinach AI is not only the most accurate AI notetaker, it also does work on your behalf:

• Sends meeting recaps and stakeholder updates over email or Slack

• Creates documentation in Notion, Google Docs or Confluence

• Adds tasks and tickets to Linear, Jira, Monday, Asana, ClickUp or Trello

• Updates CRM notes and action items on Attio, HubSpot, Salesforce or Zoho

Trusted by over 200,000 professionals. Backed by Y Combinator, Zoom & Atlassian.

Try Spinach AI

FROM THE FRONTIER

Llama 4 stampedes onto the scene

Meta didn’t even bother waiting until Monday. Over the weekend, it unveiled a new family of models, which you can try out in your browser — or on WhatsApp, Messenger, and Instagram. And if you’re developer, you can start building with it here.

Here’s a rundown of each one:

• Scout: Built for efficiency, this one’s capable of running on a single Nvidia chip. It beats open-source rivals like Google’s Gemma 3, making it “the best multimodal model in its class.” And it has an industry-leading 10M-token context window, meaning it can theoretically handle longer documents and conversations.

• Maverick: This one’s ideal for apps that involve both text and images, like a customer support bot that can analyze photo uploads. Plus, it works across 12 languages.

• Behemoth: The most powerful of the bunch, this is the model the others were distilled from. Considered “among the world’s smartest LLMs,” it outperforms GPT-4.5 and Claude 3.7 across STEM-focused benchmarks. But it’s the only one that hasn’t been released to the public yet.

The backstory: The Information reported last week that the new collection was supposed to come out sooner but failed to meet internal benchmarks — a story we’ve heard repeatedly across the industry. To keep up with the competition, Meta allegedly took cues from China’s DeepSeek, which boosted R1’s efficiency by training different parts of its model to be good at specific tasks.

THE AI ACADEMY

How to use AI agents to do competitive analysis

A strong competitive analysis requires digging through market data, understanding rival strategies, and identifying growth opportunities. You. com simplifies this process with the ARI deep research agent. Try it for yourself in just a few steps:

• Go to You. com website and enter your email address

• Enter the referral code 'SUPERHUMAN25' to gain access to ARI in just 5 minutes

• After access is given, you’ll receive an email to activate ARI. Click through the email to land on 'ARI: Advanced Research and Insights'

• Enter your research question and press Enter—no need for prompt engineering, just describe your question

Sample Prompt: Analyze the competitive landscape for ENTER COMPANY within the ENTER INDUSTRY industry. Identify key competitors (including privately held companies), their market share, and major differentiators. Highlight any recent shifts in the industry such as mergers, acquisitions, regulatory changes, or notable product innovations. Additionally, source relevant publicly available industry reports and statistics related to ENTER COMPANY and the broader ENTER INDUSTRY sector. Summarize key data points that provide insight into market dynamics, growth trends, and the evolving competitive landscape.

Wait a few minutes for AI to generate a detailed competitor report. Download it to spot market gaps, refine your strategy, and stay ahead of the competition.

PRESENTED BY TELNYX

Need Voice AI for your business? Build custom Voice AI without the complexity

Telnyx gives you full-stack, custom-built Conversational AI–backed by carrier-grade telephony, private-network speed, and owned GPUs for ultra-low latency.

Why settle for off-the-shelf solutions when you can build voice AI specifically designed for your business?

Start building Voice AI solutions with Telnyx today.

AI & TECH NEWS

Everything else you need to know today

🔬 Problem Solved: A researcher at the Brookhaven National Laboratory claims he used OpenAI’s o3-mini-high to solve a complicated physics problem that shows how phases of matter shift from one state to another.

✨ Quality Costs: Google announced it’s bringing the full version of Gemini 2.5 Pro to developers in its API. It’s the search giant’s priciest model yet — but still cheaper than Claude 3.7 and GPT-4.5.

📱 Rain Check: The Trump administration will delay its TikTok ban for another 75 days as mobile tech company AppLovin becomes the latest firm to bid on the platform.

💰 Funding: SandboxAQ, which uses quantum computing techniques to make AI more efficient, added $150M to its Series E funding round, with first-time investments from Google and Nvidia. That brings its valuation to $5.75B.

PRODUCTIVITY

5 AI Tools to Supercharge Your Productivity

✅ Clockwise: Schedule meetings and automate your calendar with AI.

✅ Readdy: Build your dream websites by talking with AI, no drag-and-drop needed.

✅ MoEngage Inform*: Save 95% Dev Bandwidth and 60% Costs as you manage Critical Alerts at Scale, with a Unified Platform for all your providers. Check it out →

✅ Tana AI: Instantly turn notes into tasks, projects, webpages or strategy docs.

✅ Runway Gen-4: Generate consistent characters, locations and objects across scenes with fully controllable media.

^{* indicates a promoted tool, if any}

PROMPT OF THE DAY

Career Growth and Development

Prompt: Help me create a personalized career growth plan based on my current role as a [your role] and my long-term goal to become a [desired position]. Identify key skills I need to develop, high-impact projects I should pursue, and networking strategies to connect with industry leaders. Include creative ways to showcase my expertise (e.g. speaking opportunities, LinkedIn content, internal initiatives) and suggest a 90-day action plan to track progress. Bonus: Recommend 2-3 people I should learn from and what I can study from their career paths.

SOCIAL SIGNALS

What’s trending on socials today

🔥 Hot Seat: For Microsoft’s 50th anniversary, current CEO Satya Nadella and former CEOs Steve Ballmer and Bill Gates sat down for an interview…with the company’s AI-powered Copilot. (The chatbot even roasted them at the end.)

💾 Memory Lane: A CSS library (a set of coding elements) inspired by Windows 98 is making the rounds online, and users are sharing the retro websites they’ve built with it. Or, check out ones inspired by the PlayStation 1 and vintage Macs.

🚀 Tight Ship: Founder Greg Isenberg explains why he thinks “the optimal startup team in 2025 is five people” — and lays out a general workflow for how it could work.

🧑‍💻 Recipe for Success: The Indie Hackers newsletter shares seven “top-tier examples” of how vibe coding is already being used to build successful SaaS products.

AI-GENERATED IMAGES

Marble Sculptures

ChatGPT Prompt: A photorealistic image of an ultra-detailed sculpture of the subject in image made of shining marble. The sculpture should display smooth and reflective marble surface, emphasizing its luster and artistic craftsmanship. The design is elegant, highlighting the beauty and depth of marble. The lighting in the image should enhance the sculpture's contours and textures, creating a visually stunning and mesmerizing effect.

Acquire new customers and drive revenue by partnering with us

Superhuman is the world’s biggest AI newsletter for businesses and professionals with 1M+ readers and 2M+ followers on socials working at the world’s leading startups and enterprises. Companies like Amazon, Hubspot, and Salesforce feature their products in Superhuman. You can learn more about partnering with us here.

🧞 Your wish is my command

Got more feedback or just want to get in touch? Reply to this email and we’ll get back to you.

Thanks for reading.

Until next time!

Zain & the Superhuman AI team