You pay for two frontier models and route almost everything to one. Both Mac apps are yours for an email, and the guide walks the verbatim prompt, the 65-check list, the job-by-job casting call, and where my own test was unequal.

I nearly wrote this up with a single winner. The app I didn’t keep is the one that showed me a question I hadn’t thought to ask.

I asked Fable 5.1 and Astra to build the same Mac app from the same short brief. Fable’s version showed me things I hadn’t thought to specify. Before seeing both, I hadn’t decided how much room I wanted the clipboard to occupy, or how much of an image I wanted to inspect before selecting it. Those decisions were easier to make with working examples.

Then the comparison got less tidy. Astra got its first version back quickly enough that I could use it and ask for changes while Fable was still working. The app I preferred contained more of my feedback. I can’t tell you how Fable would have handled the same corrections, because I didn’t give it that test.

Your favorite model gets more work, more context, and more chances to fix an answer. You learn how to describe a problem in a way it understands. Then you try a rival with that same style of request and give it less time to improve. The difference you see may include all that accumulated practice.

Astra, OpenAI’s new model, has made me reconsider how much work I give Claude. I increasingly prefer Astra for drafting, even though I still turn to Anthropic’s Fable 5.1 to think through an unsettled argument. The same project can give me a reason to use both.

Here’s what’s inside:

• Why speed changed which app I kept. A faster first build bought more rounds of corrections, and the app I use is the one that got them.

• What the app I didn’t choose was still worth. A second model can change your work without producing a single thing you keep.

• The jobs I give each one. Writing, research, spreadsheets, email, code, and 3D, with what to hand over and how to read what comes back.

• When something cheaper is the right call. The parts of a long job whose instructions are already clear, and how to tell them from the part that needs judgment.

• Shelf and Ledge. Both Mac apps, signed and notarized by Apple, ready to install.

• The guide. How I tested both apps, what broke in each, and the token and time numbers behind the comparison.

The useful part of someone else’s experience is the assignment they gave and the corrections it took to finish. Those tell me whether their result maps onto the work I have waiting.