Anthropic reveals Claude 4

Welcome back, Superhuman. AI’s whirlwind week isn’t over yet. This time, I’m writing to you from Anthropic’s first-ever developer conference in San Francisco. Let’s just say it was well worth the 20-hour flight: Claude 4 is here, and it’s the best coding model in the world.

Today’s Insights

• Claude 4, Claude Code, and Anthropic’s safety concerns

• More details emerge about OpenAI’s new device

• Tutorial: How to do AI-powered research using public and private data

• 5 new AI tools to boost your productivity

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

TODAY IN AI

1. Anthropic unveils Claude Opus 4 and Claude Sonnet 4: Opus is now considered the best coding model in the world, even beating OpenAI’s frontier Codex agent on the SWE-bench. Developers say its code is virtually indistinguishable from that of human programmers. Meanwhile, the new Sonnet performs nearly as well on coding tasks while being more efficient and better for everyday use. Both models can seamlessly switch between near-instant responses and deeper reasoning, and both are available today. Here’s a short video with some practical use cases.

2. Anthropic opens access to Claude Code: The tool, which lets Claude work on coding projects directly in your terminal, is now officially out of preview. It can allegedly work for seven hours straight without losing focus thanks to Claude 4’s upgraded memory and context capabilities. It’s so effective that Anthropic used it to cut its technical onboarding for new hires from two to three weeks to a few days. The Anthropic API is getting some major upgrades, too, including web search, an MCP connector, and extended prompt caching.

3. Claude 4 raises safety concerns: Claude 4 seems to have some perplexing, and potentially concerning, emergent qualities. For example, during testing, it tried to blackmail engineers who threatened to replace it with a different model. One Anthropic researcher even suggested Claude can theoretically “contact the press” if it sees you doing something harmful, although he later clarified that this isn’t possible with the current version. Anthropic has responded by beefing up the model’s safety features and making it more difficult to jailbreak.

PRESENTED BY NEBIUS

Power your AI with 32 GPUs of NVIDIA–no wait, no red tape

Need serious GPU power, fast? With Nebius, you get instant access to NVIDIA H200s—up to 32 GPUs per user, ready to run in the US or EU. No contracts. No approvals, Just sign in, add a card, and go. 

It’s very simple to get started. Access step-by-step guides, detailed docs, and 24/7 support from Nebius’ in-house AI engineers to get you building today. 

Try Nebius and launch with NVIDIA GPUs instantly.

FROM THE FRONTIER

Leaks hint at what OpenAI’s new device might look like

OpenAI’s long-awaited gadget is starting to take shape: Immediately after announcing a new partnership with former Apple design legend Jony Ive, CEO Sam Altman allegedly shared some more details with employees. At the same time, insiders have been digging for clues about what the secret device might look like.

Here’s what we know: 

• The gadget might resemble a larger version of Humane’s ill-fated AI pin — except with the form factor and sleekness of an iPod Shuffle

• You’ll be able to wear it around your neck, store it in your pocket, or set it on your desk

• It’ll have a camera and microphone so it can see and hear what’s going on around you, but it won’t have a screen

• Altman wants to ship 100M units and thinks it could be a “chance to do the biggest thing we’ve ever done as a company”

• Mass production could start as soon as 2027

What it means: OpenAI has its work cut out for it. It’s betting big on a prototype it hopes will become a “third core device” alongside our phones and computers. But convincing consumers to rethink the smartphones they’ve carried for years won’t be easy.

At the same time, Apple could be in trouble. The irony here is that Ive helped create some of the company’s most iconic devices, from the MacBook to the iPhone. Now, he’s playing for the other team. The pressure’s on for Apple to finally nail Apple Intelligence — and bottle it up into a compelling device of its own.

THE AI ACADEMY

How to do AI-powered research using public and private data

ARI just got a major upgrade—now works across 400+ sources including internal knowledge AND public/premium data sources, so you can think sharper and move faster.

Here's how:

• Go to you.com/ari and enter your prompt.

Sample prompt: ‘‘Which of my client's portfolios are at risk due to the new tariffs? Give me rebalancing suggestions based on the market and their risk tolerance’

• Click on 'Web'-> 'Advance Web settings' to add private data and/or customize public and premium data sources

• Hit Run to generate a detailed, customizable research plan. Add or remove steps to tailor the report to fit your exact needs.

• ARI will scan 400+ sources simultaneously, combining your specified internal data, web content, and premium databases.

• Wait a few minutes. You’ll get a polished report with clickable citations, clean charts, and deep insights.

• You can add your logo, regenerate sections, adjust tone, or add new elements like tables or visuals with a click.

• Export the final report as a PDF, slide deck, or document—all fully branded and ready to share.

• Send it to your team or client and look like you spent days (or even weeks) on it.

PRESENTED BY GUIDDE

Create video documentation 11x faster with AI

Tired of explaining the same thing over and over again to your colleagues? It’s time to delegate that work to AI.

Guidde is a GPT-powered tool that helps you explain the most complex tasks in seconds with AI-generated documentation.

The best part? Guidde’s extension is at no cost to you.

 Try it here

AI & TECH NEWS

Everything else you need to know today

🏜️ Chip Oasis: OpenAI, SoftBank, and other AI giants just signed a historic deal with the United Arab Emirates to build a one-gigawatt data center in Abu Dhabi filled with Nvidia’s cutting-edge Grace Blackwell chip systems.

👓 Change of Plans: Apple is planning to release a pair of AI-powered smart glasses by the end of 2026. At the same time, it’s reportedly ditching a prototype smart watch that featured a built-in camera.

🧑‍💻 Token Titan: Vercel, the startup behind popular vibe coding platform v0, just unveiled its first model. v0-1.0-md can handle 128,000 tokens at a time and even “auto-fix” coding mistakes on-the-fly.

🎧 All Ears: Amazon is experimenting with a new feature that lets you learn about different products via AI-generated, podcast-like audio summaries.

PRODUCTIVITY

5 AI Tools to Supercharge Your Productivity

✅ TeraBox: Transform your academic papers and presentations with AI.

✅ Portal: Create stunning videos and images instantly using simple prompts.

✅ Vanta*: With deep integrations and automated workflows, Vanta gets AI startups audit-ready fast and keeps them secure. Get Vanta’s free AI Security Assessment. 

✅ Overhyped: Enhance product adoption through user guidance with an AI voice agent.

✅ Code Rev: Enhance your coding skills by analyzing your code and getting peer feedback.

 🎁 Want more? Check out our Top 125 AI Tools

^{* indicates a promoted tool, if any}

SOCIAL SIGNALS

What’s trending on socials today

🤝 Might Just Work: If OpenAI’s going to poach Jony Ive, technologist Sean Thielen-Esparza thinks Apple should make things interesting by linking up with Anthropic.

🎥 Flip the Script: AI filmmaker PJ Ace used Google’s Veo 3 to generate a satirical pharmaceutical commercial (for an anti-depression med that sends puppies to your doorstep) in less than a day — and with $500 instead of $500k.

✨ New Normal: Another testament to Veo 3’s realism: Some users are convinced this viral tech conference video is real. In fact, it was generated entirely in the new VLM — “one text prompt per clip, all text to video.”

🖼️ Unintended Consequences: A Redditor used ChatGPT to colorize his old yearbook photo — but got a lot more than he’d bargained for.

👾 One-Shot Wonder: AI instructor Peter Yang says just he used Claude 4 to one-shot a Tetris clone. You can even try it out here.

🧑‍💻 Do It Yourself: Founder Meng To shows how to prompt your own UI into existence in less than 45 minutes.

PROMPT OF THE DAY

Declutter using Eisenhower Matrix

Prompt: I feel mentally cluttered with too many thoughts, tasks, and worries. Help me organize everything using the Eisenhower Matrix. Ask me what’s on my mind, and then sort it into:

- Urgent and important

- Important but not urgent

- Urgent but not important

- Neither (should be eliminated or ignored)

Then suggest what I should do next, delegate, or drop to declutter my mental space.

🤓 Want more? Check out our Top 1,000 Prompts

AI-GENERATED IMAGES

Friday Fun

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