I think this is a CEO vs. engineer perspective.
If you're spending millions a year on inference, investing in harnesses, routing, caching, and open models absolutely makes sense.
But for many engineers, our job isn't to become experts at model routing, it's to ship products. We work 9–5, have deadlines, and have lives outside work. Every hour spent squeezing another 20% out of token costs is an hour not spent building features.
If a Sol → Luna → Terra pipeline costs a bit more but is predictable and lets me ship faster, that's often the better tradeoff. Engineering time has a cost too.
Cataclysmic events happen and in the last one we had to depend on Noah(Vaivasvata Manu according to Hindus) to build a ship. Intraplanetary. Thanks to Elon we will have an INTERplanetary option @elonmusk
I’ve used gpt 5.6 sol , opus 5 , fable 5 , gpt 5.5 and obviously other open source models. GPT 5.6 is bang for the buck atm
Almost one shotting everything with half the token costs!
( Sure I can use an open source model too but f*** that, you need time to live life too🎾 )