they went public a few weeks ago

That's cool and all, but they are still on GLM 4.7

Which is fine for their target market. Their latest model is Kimi K2.6, available to enterprise customers. But older models become more powerful when you have time to do more reasoning. Also many applications don't need advanced models. Cerebras is making bank from all the other use cases that SOTA providers left on the table by focusing on 0-shot intelligence over speed