That has nothing to do with context window or the models at all; it's the compute capacity.
Pretty much every single AI firm had the same issue when the new model came out, it's the result of everyone jumping onboard and consuming all available compute capacity to break it for days.
Claude had the same issues when they released their newer models.
> That has nothing to do with context window or the models at all; it's the compute capacity.
Quite literally a harness feature/bug though, that it's unable to compact when reaching the max context (which seemingly both codex and Claude Code struggled with in the beginning), not something that will happen at every model release forever.
Hasn't happen for at least the two last model releases in Codex, not sure why you'd imply that.
A lot of folks are finding GLM 5.2, Kimi 3, and Deepseek just fine for their use cases.
It does not have to beat US firms, it just needs to be cheaper.
I use Deepseek v4 flash for a lot of reviewing and summarizing tasks, only used 4$ in the last two months. No dramatic drop in performance against other US models, it works for my use case. I do use GPT 5.6 Sol for other things but tried GLM 5.2 and it was good enough.
The issue was more specific to higher token burn rates, not latency.