my answer: nobody should. a benchmark just dropped tonight, fable 5.2 max scored 0.60 at $56.74 a run, opus 5.5 got 0.59 for a tenth of that. Fable is very important in my work and does things opus can't, it's just I wouldn't use it for everything and I would cap it at high personally
last time i used t3 was a legit full test of my workflow, oct 5-7 on nightly v2. liked it at first, it even taught me a few tricks. then the flaws showed up: tons of extra context on every call (pi went from 500 tokens to 26k on the first prompt), a lot of setup and adjusting to get it right, and it added a layer i had less control over without solving anything my setup didn't already handle. herdr + claude code cli gives me full control and it's still what i use today
@PovilasKorop fable 5.2 has been solving problems no other model could for me, so the per-task math is the only one that matters, it's in a league of its own on some tasks
Qwen4 is coming.
Deeper knowledge. Richer world understanding. Stronger coding and agentic capabilities.
Built for real-world tasks โ from complex coding and knowledge work to front-end builds and professional software workflows.
Qwen4-Max is now in preview on Alibabaโs Token Plan. You can also get early access to the new model on QwenWork, Qoder, Qwen APP, and QwenStudio.
Put it to the test on a real problem. Show us what you build. ๐
Token Plan: https://t.co/LOB2Jdt3Sk