Welcome to Jumble, your go-to source for AI news updates. This week, AI tokens keep getting cheaper while company bills keep exploding. Meanwhile, a GLM 5.8 just dropped and dared anyone to take it away. Let’s dive in ⬇️
In today’s newsletter:
🪙 Cheaper tokens, higher costs
👑 GLM 5.2 is the new opensource king
🪐 Musk's rocket empire snaps up a coding favorite
✊ Students stage a walkout over an AI chief
📹 Weekly Challenge: turn a doc into a slick video
📈 Tokens Got Cheaper, So Why Are AI Bills Exploding?
A single AI token costs over 90% less than it did in 2023, yet company spending on AI has doubled since late last year. Mid-2026 seems to be the end of tokenmaxxing, and the begining of strict AI budgets.
Are companies spending too much on AI?
♻️ The Jevons Paradox Returns
When something gets cheaper to use, people use far more of it. Token prices roughly halved in a year while usage jumped 450%, so the bills climb instead of falling.
🔥 Real Budgets Are Burning
Uber torched its entire annual AI budget in four months and now caps spend at $1,500 per employee a month. Even Nvidia says its compute costs outrun its salaries.
🌀 More Power, Bigger Tab
As agents take on harder jobs, each query eats more tokens, and teams keep upgrading to the newest model instead of pocketing the savings. Cheaper intelligence fuels more of it, not less.
🔓 GLM 5.2 Is the Open Model Taking on Closed US Labs
Just as Washington pulled top US models offline, GLM 5.2 shipped, and anyone can download, run, and keep, no matter what any government decides later.
Ultimately, which path wins?
🧠 A Model You Can Keep
GLM 5.2 is fully open under an MIT license with a 1 million token context, so anyone can self-host it for free (so long as you have the right hardware). Its maker, Chinese lab Zhipu, aimed it squarely at closed models that can vanish behind a policy change.
💵 Priced to Undercut
It costs a fraction of Claude and GPT, and on one marathon coding benchmark it trails Opus 4.8 by a single point. Zhipu's stock jumped as much as 48% on the news.
🌍 Open Is Pulling Ahead
While top US labs stay closed and pricey, free models you can host yourself keep closing the gap. For developers spooked by shutdowns and rising bills, that ownership is the real draw.
Weekly Scoop 🍦
🧼 Weekly Challenge: Turn a Boring Doc Into a Video
Challenge: Google Vids just made AI avatars free for everyone, so this week you turn something you already wrote into a shareable video without filming a thing.
Here's what to do:
📄 Step 1: Open vids.new and pick a source. Start a project, choose "Help me create," and either describe your video or reference an existing Slides deck with the @ symbol.
🗣️ Step 2: Let Gemini build the draft. It writes a script, lays out scenes, and adds an AI voiceover and music, all editable line by line.
🧑🎤 Step 3: Add a free avatar. Drop in one of the new AI presenters to deliver your script on camera, then export and share.
Is the AI spending boom a bubble or just the price of the future? And can an open model from Beijing really pull developers off the US giants? See you next time! 🚀
Stay informed, stay curious, and stay ahead with Jumble!
Zoe from Jumble
