This sounds counterintuitive but I’ve been doing this ever since @mvanhorn talked about purging a bunch of md files (especially lessons) and it works well for me
Grok 4.5 at $2/$6 per M tokens.
Meta with a paid tier on Muse Spark and openly competes on price.
And GLM 5.2 beating GPT-5.5 on SWE-bench Pro at 1/6 the cost open weights, MIT license.
For founders, this is a great time to be building. Your model bill is an architecture decision now.