Howdy wizards,
Today’s issue is brought to you by the awesome Context.dev, who turn any web page into structured context for your AI agents.
Here’s what’s brewing in AI.
The big thing
Someone anonymous put a serious coding model on OpenRouter, for free.
The model appeared as “Ox Alpha” last week on OpenRouter. Early community testing puts it close to Fable 5 (Anthropic’s flagship model), while spending far fewer tokens.
And it’s free (meaning no charge per token) during preview.
The catch is that the provider keeps all your prompts and completions. It won’t train on them, but it will keep them. On a coding model like this, that’s your code.
The provider says it wants to stay anonymous during the preview. Developers spent the weekend on forensics to find out who it could be.
All four anonymous models OpenRouter has hosted so far this year have turned out to be from Chinese labs.
Many think it’s Zhipu, the Chinese lab behind the GLM models, because of the connection to the zodiac sign. Microsoft is another long-shot guess.
Lots of developers started using it over the weekend, even Bloomberg picked the story up.
Why it matters
Picking the cheapest model for the job is becoming how many companies, including big ones, actually run AI.
And models like Ox Alpha can probably score you a big discount on your AI bill.
But at what price? Your code leaves the building, then gets kept, and you don’t know by whom, in which country, or under whose law.
IN PARTNERSHIP WITH CONTEXT.DEV
Your agent is only as useful as the data it can access. Context.dev gives AI agents and LLM-powered apps a single API for real-time web data: scraping, crawling, structured extraction and brand intelligence.
Skip the brittle scraping stack and give your product clean, usable web data with one integration.
NEWS NEWS NEWS ❦ NEWS NEWS NEWS
All the small things
New tools & product features
This case study is all over the internet right now: Asana’s five-year code cleanup took OpenAI’s Codex two weeks. The job was removing Enzyme, an outdated testing system, from its codebase. They used a five sentence prompt and it took 1.5 weeks. It was supposed to take 5 years and cost $6M, but the run with Codex ended up costing about $12,000, per OpenAI. Can you smell the BS in this story?
Industry moves
Nvidia is paying $6B to license Poolside’s coding models. The logic is this: companies currently rent their models from OpenAI and Anthropic. Nvidia sells chips. If Nvidia has a model of its own, companies might start buying chips and running it themselves. Also, this is not an acquisition: Poolside carries on, minus 109 staff that Nvidia hired.
California’s No Robo Bosses Act is back. It’s a law that bans firing or disciplining workers purely based on AI’s decision; there needs to be human oversight and independent verification when AI helps decide. It’s an interesting parallel to what’s going on in Europe where it’s supposedly illegal to reject CVs based on an algorithm without a human in the loop; though they’ve found easy loopholes around that.
Research
Claude designed proteins that actually stick. One in four of its designs worked (one in eight is the normal rate).
Researchers find over a third of all web pages published since ChatGPT have AI writing on them. I like that we’re getting empirical about this, though I think just about no one is surprised at the result.
❦
You are a delight.
What's your verdict on today's email?
See which AI use cases are paying off with Context Windows Pro
Most companies pick AI use cases by brainstorming internally. 90% of those initiatives fail.
I’ve created Context Windows just so you can pick the winners.
🟦 Find high-performing use cases from 2,000+ companies at contextwindows.ai, or book a demo with me
Disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes work with sponsors and may earn a commission if you buy something through a link in here. If you choose to click, subscribe, or buy through any of them, THANK YOU – it will make it possible for me to continue to do this.