In partnership with
OpenAI says this is "the agi era." you?
This is our feed in the last 2 weeks summed up. 8 labs, all "introducing," and the meme up there doesn't even have the big one.
I do this full time and even I tapped out at some point, but I was reeled back in thanks to OpenAI’s launch last night.
“Introducing” Astra
OpenAI shipped GPT-6 Astra last night and called it "the world's most intelligent and aligned model." Every lab writes that line.
The launch was messy, the blog post went up, vanished, came back. ChatGPT started throwing errors while the other models were down too.
Half the people who sprinted to try it got told they weren't on the list. The smartest model alive, and its own website tapped out.
The ones who did get in lost their minds.
One tester spent the night building a city out of ASCII text, seven explorable little worlds, and a town simulation with jobs and elections.
Another said it's the best model they’ve ever touched, and one-shotted a coding job that had beaten 2 other models for months.
How to get it: rolling out now, cyber-programme and enterprise first, then paid ChatGPT over the next few days.
In ChatGPT it shows up as "GPT-6 Pro" on the Pro plan. OpenAI's own help page says Plus doesn't get it, so if you're on Plus, sit tight.
Try this: stop typing questions at it, give it a job. "Take this PDF invoice, pull the line items into a sheet, email me the total." If you're still using it like a chatbot, you're wasting it.
The rest
Claude got cheaper. Fable 5.1 cut cache reads 75%, so normal use runs about 25% less for the same model. If you pay for Claude, that's free money.
Microsoft made transcription nearly free. MAI-Transcribe-2 handles 60 languages with speaker labels and word-level timing at $0.10 an hour, roughly 72% cheaper than its last version.
Abu Dhabi open-sourced a whole model fleet. K2 Horizon is 6 models from 0.9B to 375B with the weights, code and training data all free, and the small ones run on a phone.
Apple's turn is Wednesday, 9 Sept. New CEO, a first foldable iPhone, and the Gemini-powered Siri they've promised for a year.
The money and rules
And in one breath: Adobe named a new CEO, and Nvidia shipped a free tool to run AI models across the spare PCs in your house.
Tools worth stealing this week
Sourclip: a Chrome extension that turns Gemini Notebook (the old NotebookLM) into a real research workspace, one-click capturing webpages, YouTube, PDFs and whole AI chats into your notebooks.
TrustedRouter: one login for 600+ models across 80+ providers, wired so the router never sees your prompts. Run Astra, Fable and a cheap Qwen side by side without juggling tabs.
Tovel: records your calls and meetings, then turns each one into action items, follow-ups and CRM updates. For the notes-to-actions bit you always skip after a client call.
This week's cheatsheet
The models can do your work now, the catch is knowing what to ask.
So this week's cheat sheet is 18 prompts that turn a mess of documents, transcripts and open tabs into briefs, tables and emails.
My take
Everyone shipped, the scary versions got locked up, and the one you can touch got cheaper and better. Ignore the leaderboard wars.
Pick one job this weekend, a week of research or a first-draft deck, and hand the whole thing to Astra, Fable or a cheap Qwen.
Watch it work without you.
I’ll cover them one edition at a time, so if you’ve anything specific you’d like me to test or try, reply and let me know.
Until next time,
Vaibhav 🤝🏻
If you read till here, you might find this interesting
#AD 1
No follow-up questions required
Every sales leader knows the feeling. You walk into a pipeline review with a number you believe in, and twenty minutes later, you're defending every line item to a CEO who just wants to know what's actually going to close.
HubSpot Sales Hub ends that conversation. Every deal, every rep's activity, and every buyer signal are all in one place and updated automatically. So your forecast is built on what's actually happening. And when you present that number, you can stand behind it.
#AD 2
The best voice models, now with full orchestration. Build real-time voice and chat agents on one low-latency stack: any LLM, your tools and knowledge, testing, Guardrails, and omnichannel deployment.
