What Happens When a Model Disappears?

Hi friend,

The big story of the past few days was obviously Fable 5. I tested it and found it genuinely extraordinary, but my conclusion was still that it is definitely not the new default model most users should switch to.​

​
What happened next has already been covered thoroughly. The US administration, citing national security concerns, barred Anthropic from offering access to the model to non-US citizens. Anthropic responded by pulling access more broadly.

​
I’m not going to comment much on the politics of this here. Many people smarter than me have already written about it at length. If I had to point to one source I usually like to read to understand these topics better, it would be Stratechery.

​
What the whole odyssey did spark in me, though, was a more practical question:

If a model can just disappear like that, what else might happen? And if more of my work starts to depend on external intelligence in the form of AI models and tools, how should I design my workflows so that one missing model, provider, or tool does not break everything?

​
I don’t think the answer is paranoia.
I’m not interested in overengineering every workflow just to be extra safe. That usually creates more complexity than it solves. But I also think it is risky to build too much of your work around one model provider, one specific model, or one tightly integrated tool.

​
This is especially true if you run a business that is starting to rely on AI not just as a nice productivity boost, but as part of how work actually gets done.

​
Some people have argued that the whole Fable 5 story was the best marketing open source AI models could have asked for. I think that is right.
​
For the sake of focus, I have mostly used frontier models from the big labs in my daily work. But right now, I’m starting to test how well open source models actually work for my own use cases.
​
The signals are much more promising than they were six months ago. This week’s release of z.ai’s GLM 5.2 pushed that point even further, with benchmark results that appear surprisingly close to the frontier models.
​
So over the next few days, I’ll be testing more open source models in real workflows and see how they stack up against the frontier competition. I’ll keep you posted!

Thanks for reading,
Robert

Headlines

• Vercel launches eve - a file based agent framework. Think of it as the Next.js for agents. Looks promising!
• Popular Codex features like Computer use are finally rolling out in the EU

dominik kundel @dkundel For our European friends 🎉 we are rolling out this week Codex for Chrome, Computer Use, Memories, and Chronicle in EU, EEA and UK. Here's a recap on how you can use Codex with Chrome and some more tips below 🧵 4:35 PM • Jun 16, 2026 10 Retweets 195 Likes Read 19 replies

​

• ​OpenRouter's Model fusion is an interesting approach using routing and parallel model invocation to get better results at lower prices.