Welcome back! Alibaba priced its biggest model at half what OpenAI charges and said the weights go free next week. An OpenAI model nobody can use yet cleared ten math problems that had sat untouched for decades, YouTube spent months deleting slop farms, and Altman took his next model to Washington, meaning a launch is expected soon.
In today's Generative AI Newsletter:
Alibaba: What are you getting for free next week?
OpenAI: What did $2,000 of tokens prove?
YouTube: How to survive the purge?
Astra: Who has seen OpenAI's next model?
Alibaba undercut OpenAI by half
Alibaba previewed Qwen3.8-Max two weeks ago with no price and no benchmarks. Today it went on sale, and the numbers are impressive.
Qwen charges $2 per million input tokens and $6 for output. OpenAI cut Terra to $2 and $12 on Friday, so Qwen launched at half the output price three days later.
The open weights for the full 2.4 trillion parameter model go up next week.
The smaller Qwen3.8-27B goes open too, which is the one that runs on hardware you can actually buy.
Alibaba's headline claim is a model that codes alone for more than ten days, from an empty folder to something shipped, with the whole project history published on GitHub.
Take the rankings carefully. Alibaba says it sits second only to Fable 5, but this is based on its own internal evals. If it’s actually true we should all reconsider our AI strategy.
10x the context. Half the time.
Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.
Time's running out on this opportunity
GenAI Works reaches 14 million people, works with 300+ AI brands and counts Nvidia, Oracle, Google, IBM and Notion as clients. Revenue grew 2.5x last year.
We’re building applications that will expedite our growth, and this is why we’re raising. You get an early opportunity to own a piece of the future. $2.72 million already in.
Start from $1,000 and get up to 22% bonus shares in Phase 2.
An unreleased model solved 10 problems that stood for decades
OpenAI published ten new results on Saturday, on problems that had seen no progress for at least ten years and in most cases far longer.
The work came from an internal version of Astra, a model nobody outside OpenAI can touch.
It established that non-sofic groups exist, settling a central question in group theory.
It disproved Connes's rigidity conjecture.
It resolved three problems Paul Erdős left behind, numbers 183, 146 and 180.
The tokens needed to find all ten solutions would run about $2,000 at Sol's API rates.
Humans wrote the papers. The model then formalized every proof in Lean, so a computer can check the logic line by line, and OpenAI put the certificates on GitHub.
This is a revolution in science. Are we at the age of AI being responsible for scientific breakthroughs? Looks like it’ll be the case for the years to come.
YouTube deleted 130,000 channels for AI slop
Over six months, YouTube's automated system terminated 130,000 channels.
It's called the Scalable Cluster Termination System, and it hunts for coordination rather than judging videos one at a time.
That's how it kills a hundred channels in one move.
AI-assisted work with real human input and proper disclosure keeps its monetization. What gets killed is templated output with no human contribution behind it.
YouTube says fewer than 1% of appeals get overturned, though creators caught by mistake have been loud about false positives.
This is an extension of the spam and clickbait systems YouTube already ran, rather than something new.
If you make videos, disclosure is now the thing standing between you and an automated ban.
Altman showed Astra to regulators before anyone else
OpenAI is testing a model family called Astra, a new class sitting alongside Sol, Terra and Luna.
Astra is built for work that runs long. Multiple agents collaborating for hours or days on problems that normally need a team of people.
Sam Altman demoed it to policymakers and regulators in Washington last week, before any public release.
Nobody outside OpenAI knows what it ships as. It could arrive as GPT-6, or as GPT-5.7 if OpenAI keeps stretching the current series.
Traders on Polymarket put the odds of a release by the end of next month at 56%.
The ten math proofs above came from an internal version of this one.
Tool of the Day: Pipedrive AI
Pipedrive is a sales CRM with an AI layer that reads your pipeline and tells you which deals to chase, writes and summarizes your emails, and builds reports from a plain sentence. The reporting, marketplace search and app recommendations sit on every plan instead of the top tier.
Try this yourself:
Start the 14-day trial. No card required.
Import real deals so the AI has a pipeline to read.
Open Insights and type the report you want in plain English.
Run the AI email writer on one lead and compare it to yours.
Who it's for: sales teams spending more time updating a CRM than selling.
Everything else you shouldn't miss
GPT-5.4 disappears from ChatGPT on August 31: It stays alive on the API and in Codex sessions authenticated with a key, so anyone pinned to it in the app has four weeks to move.
Fake AI doctors are selling you supplements: AI-generated physicians are pulling millions of views on TikTok pushing unapproved supplements and disproven cancer myths, including that deodorant and sleeping next to your phone cause it.
The researcher who wanted "rapid RSI and human disempowerment" left OpenAI: He's off to found a startup, which is either reassuring or the opposite depending on what he builds.
Nvidia's CEO says the six-figure AI jobs are in plumbing: Data center construction workers averaged $81,800 in early 2026, about 32% above comparable commercial builds, and McKinsey puts the US short 130,000 electricians by 2030.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
