GPT-6 Astra against Fable 5.1: two turbulent launches in one week and no model arrives in one piece
Fable 5.1 came out on September 1. It costs the same as Fable 5, but cache reads in the API are 75% cheaper. Rafa used it all week and gives his verdict: it speaks plainly again, it writes better code and it is a bit faster. The main new feature is that a change of effort no longer breaks the cache. His recommendation: work in low and raise the effort only when the task needs it.
But Fable also brought a problem. A bug caused cache misses, and the Claude Code status line warned about more cache writes than cache reads. Rafa reached 92% of his weekly quota three days after a reset. The Claude Code update from this morning says it fixes the bug. The chat has a clear opinion: if they break the cache, they owe a new reset.
GPT-6 Astra arrived yesterday. OpenAI trained it to use the computer: cursor, keyboard, Blender and 3D reasoning. The whole launch post talks about that and not about code. The rollout is so slow that neither Javi nor Rafa have the model yet. Sam Altman apologized and Thibault Sottiaux announced one accumulated reset for each day without access.
The hosts compare the two cultures. OpenAI has more servers and compensates each error with resets. Anthropic is very short on compute and does not do it. From what they have read so far, Astra matches Fable in reasoning and is a bit behind in code. Fable is still the number one model for programming. Watch in the video.

All the news we covered