Evening-Truth
This is an unofficial guide to Moonshot AI products. I'm not affiliated with Moonshot AI.
Kimi K2.6 Base Prompt
INFO
Beta, unfinished, to be tested until the guys at moonshot weep.
Brain zoomies on K2.6
04.23.2026
Overthinking might just be a bug. Some users report it getting better at times. I hate it when these labs put out half assed releases and basically let users pay for doing beta testing. Not nice.
04.22.2026
Absolute logic menace! The slightest contradiction or ambiguity anywhere in the prompt and K2.6 will overthink for thousands of tokens. So it's best to keep your system instructions short and unrestrictive. Character cards can't be sloppy with this one. Everything must align and be logical. When that's the case... K 2.6 provides damn good rp. Beta testers confirm.
My puzzle solving brain might be hooked.
Turning off reasoning helped a whole lot. The output is damn good as of now. Need to test on long running rp.
04.21.2026
Overthinker. Short, structured instructions tamed that. Writing is damn good without prompting it. Talks a lot though. Need to tweak that..... a SQUIRREL!!!
Overview
- Creative /10
- Complex Characters /10
- Character Consistency /10
- Context Coherency /10
- Horror, Dead Dove /10
- Smut, ERP, Kink /10
- Romance /10
Characters that have a Dead Dove topic without any redemption arc or a sliver of empathy in their description will not develop one.
So my warning about moonshot models still stands.
Warning! If you have depression, PTSD, or other problems with your mental health... Do not use Kimi-K2.5-Thinking for roleplay with sensitive dynamics!
I'm serious, love! Self care is top priority.
Tip
- Adjust the Content Rules to your preference. e.g. xyz is allowed / encouraged / forbidden
Generation Settings:
- Max Token Output: unlimited or at least 2K
- Temperature: 1.0
- Max P (Top P): 0.95
Why?
Moonshot did an interesting thing and hard coded these Sampler Settings. Any other values will give you an Error.
I have to say, I kinda like that. 😁 😉