OpenAI launched strongest new models

Welcome back, Superhuman. OpenAI launched its strongest model yet, but only to trusted partners at the government’s request. CEO Sam Altman called this slow rollout reasonable, but not optimal. Jokes about the permanent underclass are starting to feel uncomfortably close to reality.

Today: Google rations Meta’s access to Gemini, how to turn Claude Cowork into your personalized assistant, and get the latest prompts and trending social posts.

1. OpenAI launches GPT-5.6 series to a limited audience: The GPT-5.6 model series includes Sol, OpenAI’s strongest model yet, Terra, Sol’s more affordable daily driver, and Luna, a budget option. At the U.S. government's request, OpenAI is opening to a limited circle of partners first to ensure safety. This mirrors the treatment of Anthropic’s Mythos, which is now back in limited preview, with Fable rumored to be released this week as well.

2. New Chinese models close the gap with America’s frontier labs: Tulongfeng is a new model from Chinese cybersecurity company 360 Security Technology that its founder billed as a "Chinese version of Mythos” when it comes to identifying cyber bugs. The recent release of Z.ai’s GLM-5.2 has also put pressure on US frontier labs by rivaling top model performance at a fraction of the price. Read more.

3. Google rations Meta's access to Gemini: Google has capped Meta's use of its Gemini AI models, citing a lack of compute capacity. Google is so compute constrained that it even agreed to pay SpaceX $920 million a month to access 110,000 Nvidia GPUs. The situation highlights how the AI infrastructure buildout has not kept pace with consumption, and that the ongoing compute shortage is starting to reshape corporate relationships.

Your Agent. Your Model. Your Rules. 

Strands Agents is the open source agent harness SDK, powering everything from production backends to physical robots to code that writes itself. Build production-ready agents in a few lines of code, with any model, anywhere.

Build your first agent harness.

Memory chips surge. Semiconductor and memory chip firms haven’t been able to lose this year. Last week, Micron (+296% YTD) reported that quarterly revenue quadrupled to nearly $42B, SanDisk (+780% YTD) saw last quarter's sales jump 251% to $6B, and SK Hynix (+294% YTD) has grown so quickly it’s now South Korea’s most valuable company. But these companies have been the exception in the stock market lately — with many marquee tech names far below their all-time highs.

Software’s bad year continues. Earlier this year, fears that AI would disrupt enterprise software (known as the SaaSpocalypse) sent names like Oracle, Salesforce, and ServiceNow cratering. Months later, all three are still down more than 50% from their all-time highs, although the White House’s move to restrict OpenAI's new GPT-5.6 models triggered a sharp increase last Friday.

Big Tech stumbles. Meta (-29%), Amazon (-15%), Alphabet (-16%), and Microsoft (-28%) have all slipped from their highs in recent weeks. One catalyst: high spending. These companies plan to sink roughly $725B into AI infrastructure this year, with a growing part of the bill being funded by debt. Investors have pulled back until there’s more proof this spending is paying off.

But the market is green. For all the red, the S&P 500, Dow Jones, and Nasdaq are still up 7 to 9% on the year — far far from recession territory. Still, expect volatility to be a key theme for the remainder of 2026, as a single model launch, feature, or leaked memo can swing markets hard in either direction.

Sample Prompt: I want to create a custom Al assistant settings document I can paste into my AI tool (like Claude Cowork, a custom GPT system prompt, or similar). First, assess what you already know about me from our conversation history. If you have enough to write something specific and opinionated, go ahead and draft it. If there are gaps that would make the output generic, ask me only the questions you actually need answered - don't interview me for information you already have. Cover: how I like to work, my communication style and tone preferences, what I'm building or managing, how I want the AI to handle ambiguity or pushback, what kinds of decisions or actions require my explicit approval before proceeding, and any recurring workflows or contexts the AI should know upfront. The output should be a formatted settings doc I can paste directly in - specific to me, not boilerplate. Try to condense because this is loaded every time.

SmartBear Reflect is an AI-powered, codeless test automation platform that turns plain-English test steps into automated actions—no coding required. 

Try it at no cost.

🦾 Skill Upgrading: This viral post outlines how to build a personal productivity system around Claude Cowork. Once it’s built, the system streamlines your daily production while you handle decision-making. Setup and prompts included (4.5M views).

👤 Historical Figures: See what Socrates, Mozart, Cleopatra, and other historical figures might look like today, according to these viral AI images (1.5K upvotes).

📉 Token Efficiency: CEO Brian Armstrong shared how Coinbase keeps AI spend flat, despite using “exponentially” more tokens. His top strategies: defaulting to open-weight models, improving routing, and keeping context lean (3M views).

🧑‍💻 PDF Viewer: A Redditor vibecoded a simple tool that makes navigating PDFs on a computer way easier. Commenters say it’s a genius idea that should’ve become the default years ago. Try it here.

🥵 Heat Wave: How bad is the historic “Heat Dome” enveloping Europe right now? These AI-generated videos capture just how brutal the heat feels (1M views).

^{* indicates a promoted tool, if any}

Anthropic called out Alibaba for “brazenly” extracting Claude’s abilities and called on Congress to take action. The AI lab says this was the biggest distillation attack yet, dwarfing a similar campaign earlier this year.

Grow customers & revenue: Join companies like Amazon, HubSpot, and Salesforce. Showcase your product to our 1M+ readers and 2M+ followers on socials. Get in touch.

Until next time — Zain, Theodore, & the Superhuman AI team.