Quality using Kimi K2.6 via kimi coding plan
Hi,
After the "Claudeocalypse" last month I have been testing several models in order to find the best quality/price model to pay for when using Hermes.
I found GLM5.1 quite good, but when I tried to upgrade my membership to the next tier, I saw what was a 30$ subscription had jumped to 64.80$, and the next tier was 144$.
I then checked for a fallback model that could be similar enough and the one I ended purchasing was the 19$ "Kimi Code" plan, which uses their Kimi K2.6 model. All benchmarks suggested that it would be similar or better than GLM5.1.
The result has been...disappointing. I can feel a significant downgrade in quality regarding GLM. The model feels dumber. Does not plan, does not check, forgets data and tools to use. Goes in circles trying to fix what GLM could fix straight away, etc...
I have checked if I had any problems with my setup or configuration, maybe I was using an older model, maybe the thinking budget was off, but I can not find anything.
It feels like a model from 2024-2025.
Do any of you have experienced the same?
Thank you