i thought the codex models didn't have taste until i switcehd to 5.3 codex spark for a bit.
produces way less weird code that technically works but is ugly af
@HeroOfKVcache agreed, using it on medium thinking for medium complexity tasks felt like a sweet spot where I wasnβt having to break up a task across multiple sessions like I would with opus
@DennisonBertram The other issue is each model from zai has different concurrent request limits. But yeah I wouldnβt use them for heavy multiagent workflows even though the models are good