You Blinked. OpenAI Launched Everything

Welcome back apprentices! 👋

OpenAI just had one of those weeks — the kind that makes your roadmap feel outdated, your chatbot look like a Tamagotchi, and your CTO suspiciously quiet on Slack.

No spoilers yet, but let’s just say this wasn’t your average “here’s a shiny new model” release. 

From brainy upgrades to budget-friendly twists, surprise product hints, and a casual billion-dollar power move or two... the team at OpenAI is clearly speedrunning the future.

Strap in — we’re breaking down everything they just launched (and maybe leaked).

In today's email

• 12 Ex-OpenAI Staff Defend Nonprofit Roots

• Google's New AI Speaks Dolphin — No, Seriously

• NVIDIA to Build AI Supers in Texas, Y’all

• A Casual Week at OpenAI

• Even more AI magic

• Test the prompt (NEW)

_{Read Time: 4 minutes}

Quick News 

🧑‍⚖️ Twelve former OpenAI employees just jumped into Elon Musk’s lawsuit like it’s the group chat that finally spilled. In a spicy amicus brief, they claim OpenAI went from “saving humanity” to “profit, please” after cozying up with Microsoft. With Harvard’s Lawrence Lessig backing the filing, the whole thing feels like Silicon Valley’s version of a messy tech divorce—complete with old emails, founding ideals, and AI ethics on trial.

🐬 Google just announced DolphinGemma, an AI model trained to analyze and generate dolphin vocalizations — it's basically ChatGPT for Flipper. Developed with Georgia Tech and powered by Google’s Gemma models + audio tech, it was trained on decades of underwater squeaks, clicks, and whistles from the Wild Dolphin Project. The AI mimics the structure of language models, predicting what dolphins "say" next, and even powers a Pixel 9-based underwater chat device — yes, real-time dolphin interaction is officially a thing now.

🏭 NVIDIA is bringing its AI supercomputer empire home, kicking off domestic production of Blackwell chips and AI systems in Arizona and Texas. With over 1 million square feet of chip-fueled ambition and partners like Foxconn and Wistron in the mix, it’s aiming to crank out up to $500 billion in U.S.-made AI infrastructure over the next four years. Basically, Jensen Huang just put on a hard hat and said, “Let’s build the brains of AI right here.”

Together with Writer

You’ve heard the hype. Now it’s time for results

After two years of siloed experiments, proofs of concept that fail to scale, and disappointing ROI, most enterprises are stuck. AI isn't transforming their organizations — it’s adding complexity, friction, and frustration.

But Writer customers are seeing a positive impact across their companies. Our end-to-end approach is delivering adoption and ROI at scale. Now, we’re applying that same platform and technology to bring agentic AI to the enterprise.

This isn’t just another hype train that doesn’t deliver. The AI you were promised is finally here — and it’s going to change the way enterprises operate.

See real agentic workflows in action, hear success stories from our beta testers, and learn how to align your IT and business teams.

Watch the product showcase now

OpenAI 
From AGI to IG? A Week of Smarter Bots and Surprise Social Plays.

This week, OpenAI essentially said: “Why not ship everything at once?” In classic Sam Altman style, the company launched three new models, toyed with taking on social media, tossed some open-source love at terminal geeks, and casually considered a $3B acquisition — oh, and a global supercomputer network for good measure.

Here’s your full download... before they drop another surprise in, say, 12 hours.

GPT-4.1: Now With a Million Tokens and Fewer Existential Crises

If GPT-4o was the cool cousin, GPT-4.1 is the older sibling who went to Stanford, builds React apps in their sleep, and now charges less for it. It's joined by 4.1 mini and 4.1 nano, two leaner, faster sidekicks designed to run laps around previous models without melting your API budget.

Key flexes:

• Up to 1 million tokens of context — enough to stuff in your whole startup repo and a novella about it
  

• 26% cheaper than GPT-4o (the economic AI we love to see)
  

• Evaluators preferred its UIs 80% of the time, which is basically a win on Hot or Not for code

So yes, the name might sound like a firmware patch, but it’s more like a dev-focused brain with photographic memory. And it’s API-only, because ChatGPT is for the normies now.

Flex Processing: Half-Price AI, If You Can Wait a Bit

In a move that blends affordability with a touch of “we’ll get to it soon,” OpenAI launched Flex Processing — a discounted API option for o3 and o4-mini that slashes costs by 50% in exchange for slower, non-guaranteed response times. Perfect for non-production tasks like bulk data enrichment, model evaluation, or asynchronous jobs, Flex is OpenAI’s way of letting you run state-of-the-art models without setting your wallet on fire.

Here’s what that looks like:

• o3 drops to $5/M input tokens and $20/M output (down from $10/$40)
  

• o4-mini comes in at $0.55/$2.20, sliced from $1.10/$4.40
  

The catch? Responses are slower and resources aren't always available on demand — so it’s best for background jobs, not real-time apps. Also, ID verification is now required to access o3 (and some features like streaming and reasoning summaries), especially for users in lower usage tiers — part of OpenAI’s push to keep the bad actors out.

Flex drops just as competitors like Google’s Gemini 2.5 Flash turn up the heat on low-cost inference, showing that OpenAI’s not just competing with smarter models — it’s also tweaking the economics of how and when you run them.

o3 and o4-mini: Models That Use Tools and Think in Pictures

Next up: o3 and o4-mini, the models that don’t just reason — they multi-tool, multitask, and low-key solve science problems while you’re still writing the prompt. They’re like if your whiteboard brainstorm had a physics PhD and an art degree.

They can:

• Use all ChatGPT tools mid-thought (Python, DALL·E, browser—bring it on)
  

• "Think with images", meaning they don’t just see — they interpret, generate, and remix visuals
  

• Solve elite math benchmarks like AIME 2025, which most humans can’t even pronounce
  

• Run Codex CLI, the command-line copilot we didn’t know we needed (but now absolutely do)

Greg Brockman called this a “GPT-4-level qualitative step into the future.” 

Translation: it’s basically Step 4 on the AGI ladder, and these models just took it with toolbelts on.

Social Media, But Make It OpenAI

Because shipping all that wasn’t enough, The Verge reports OpenAI might also be building a social network. (Yes, seriously.)

Details are hush-hush, but here’s what leaked:

• It’s internally prototyped
  

• It showcases AI-generated images and creative content (imagine Instagram, but trained on your weirdest DALL·E prompt history)
  

• It’s being pitched quietly to outsiders
  

• Altman’s original comment? “Fine, maybe we’ll build a social app.” (Reader, he did.)

If this goes live, OpenAI won’t just train on your tweets — it might own the whole platform. And let’s be honest: it already has the world’s most talkative user base.

Big Money Moves: Windsurf & World-Dominating Compute Projects

And while all this innovation was happening, OpenAI casually dropped some billionaire energy on the side.

According to Bloomberg  it’s in talks to acquire Windsurf — an enterprise AI infra company — for $3 billion. Meanwhile, FT reports OpenAI is expanding its Stargate supercomputer project globally, eyeing partners across the Middle East.

Yes, they’re building literal AI fortresses on multiple continents. No big deal.

So... What’s the Deal?

This week wasn’t just a release sprint — it was a preview of OpenAI’s full-stack strategy in motion:

• GPT-4.1 is the developer brain
  

• o3 is the research brain
  

• Codex CLI is the command-line sidekick
  

• The maybe-social app is the dopamine drip
  

• Flex Processing is the wallet-friendly lever that makes scaled usage actually doable
  

• Windsurf brings enterprise-grade infrastructure in-house
  

• And Stargate? That’s still the intergalactic GPU mothership

TL;DR

OpenAI isn’t just powering your chatbot anymore — it’s quietly becoming the underlying operating system for building, thinking, automating, creating, and (yes) doom scrolling. It's building models, tools, pricing knobs, data centers, and now maybe a place for you to post your AI-generated vacation pics.

If your company roadmap still involves quarterly standups and sticky notes, remember: OpenAI launched three new models, a command-line agent, a pricing strategy, and possibly a social platform in one week — and still made time to go datacenter shopping.

Now breathe. And seriously, update your strategy deck.

Help Your Friends Level Up! 🔥

Hey, you didn’t get all this info for nothing — share it! If you know someone who’s diving into AI, help them stay in the loop with this week’s updates.

Sharing is a win-win! Send this to a friend who’s all about tech, and you’ll win a little surprise 👀

Even Quicker News 

🩺 An AI called ULTR-AI just out-diagnosed human experts in spotting TB — scoring 91% sensitivity and 85% specificity using a pocket-sized ultrasound in rural Benin. No sputum, no lab — just a smartphone, some machine smarts, and a radiologist’s worst nightmare.

🛡️ NATO just bought Palantir’s AI war brain and plans to plug it into real-world missions within 30 days — because nothing says “trust fall” like handing targeting decisions to a large language model. Let’s hope it doesn’t autocomplete a battlefield.

🪸 ByteDance’s Seaweed just outperformed Sora, Veo, and the rest of the video AI big shots — with just 7B parameters and way less compute, proving you don’t need a GPU bonfire to make cinematic magic. It does text-to-video, lip syncs like a K-pop star, and tells multi-shot stories without breaking a silicon sweat.

Today’s Toolbox

📸 Microsoft’s new Recall feature takes screen snapshots every few seconds so your PC can remember everything you forgot — including that one weird tab you definitely meant to close. It’s like having a photographic memory... with a mild privacy crisis baked in.

🎨 Canva just leveled up from “drag-and-drop” to “type-and-magic,” letting you generate images, apps, and charts like a design sorcerer with an AI wand. Somewhere, Adobe just spilled its coffee.

✨ Kling just dropped KLING 2.0 and KOLORS 2.0 — AI so good it might direct your next indie film and design the poster. It edits, animates, and even follows your prompt like it actually reads the brief (looking at you, Western models).

🧪 Test the Prompt

A playground for your imagination (and low-key prompt skills).

Each send, we give you a customizable DALL·E prompt inspired by a real-world use case — something that could help you in your business or job if you wanted to use it that way. But it’s also just a fun creative experiment.

You tweak it, run it, and send us your favorite. We pick one winner to feature in the next issue.

Bonus: you’re secretly getting better at prompt design. 🤫

👑 The winner is…

Last week, we challenged you to test GPT-4o’s visual generation skills with this prompt.

We could’t decide, so this time we want you to choose the winner!

Congrats to both Andrew (creation 1) and Tamara (creation 2)!🥳

Want to be featured next? Keep those generations coming!

🎨 Prompt:

“Illustrate your brain as a chaotic office run by tiny assistants shaped like [household object], located inside a [weird workplace setting].
Style: colorful isometric with labeled details.”

We’ll be featuring the best generations in our next newsletter!

_FEEDBACK