Insane degradation since model change

What version of Kimi Code CLI is running?

kimi, version 1.43.0

Which open platform/subscription were you using?

Moderato

Which model were you using?

k2.6

What platform is your computer?

No response

What issue are you seeing?

On ver 1.30.0 with kimi-for-coding, tasks just worked. It got it right the first time — no loops, no getting stuck, no drifting off into irrelevant exploration.
On 1.41.0 with K2.6, the same prompts and same codebase turn into a mess. It starts chasing dead ends, piles up garbage context, and forgets the constraints I set at the start. Quality is night and day.

it was the best, period. i would take it over anything. now i am probably going to have to switch which bums me out. please fix, please.

What steps can reproduce the bug?

its easily reproducable , it's any session now. its constant. i have version 1.30 with kimi-for-coding model on my other pc and its night and day. i cant believe how bad it is now in comparison.

What is the expected behavior?

version 1.30 with kimi-for-coding model is chefs kiss. at this point it should be better than that, not worse.

Additional information

No response

No files, tests, or entry points are named. Start by reproducing the behavior in Kimi Code CLI 1.43.0 with K2.6 and compare equivalent sessions against version 1.30.0 with kimi-for-coding; done would require identifying and correcting the regression so comparable prompts no longer loop, drift, or lose constraints.