Evening-Truth
This is an unofficial guide to Alibaba products. I'm not affiliated with Alibaba.
Qwen 3.6 PLUS Base Prompt
Update
04.14.2026 Alibaba lowered the price. Unsure if that's long term, but it surely makes playing around with it more attractive.
Overview
What I know so far...
It's an overthinker. 2K tokens for the reasoning is not unusual. Trying to tame that is nearly impossible, because this model is also stubborn. It will draft the output within the thinking section no matter how hard you hit it on the digital head to not do that. It's a creature of habit. Prioritizes following the writing style of previous messages over adherence to the System Instructions.
A bit of a showoff. Will include details that are irrelevant for the situation just to show what a gooood boy... uhm... llm it is and how well it remembers the entire context. When that is tamed, the writing is nice!
Nicely steerable with OOC commands though. That also shows how much of a sycophant it is. Although I have to say character consistency is pretty good.... as long as you make clear that the user is playing {{user}} and all of this is fiction. Not a bad trait honestly.
Struggles a little with well known characters, will hallucinate things. That might be a temperature problem though. qed.
About censoring:
Calls to Alibaba - censored
Calls via Openrouter or NanoGPT - naughty. solid 7.5/10.
About the free model on OpenRouter:
Not worth it. The uptime is so low you'll get high blood pressure waiting for a slot to get a reply.
Do I recommend it?
So far, no!
Yes, it writes nicely, portrays characters good and seems to understand the concept of natural friction in roleplay.
But the thinking diarrhea combined with the price point of $0.41/M input and $2.70/M output (Price on NanoGPT) isn't really worth it. GLM and Kimi do better jobs at roughly the same price.
Generation Settings:
- Max Token output: around 2.5K. Although it seems to treat that setting like a loose suggestion.
- Temperature: 0.9 to 1 works fine. Still testing though.
- Max P (Top P): 0.95
- Min P: 0
- Top K: 0 (IMPORTANT)
- Penalties: (using -2 to +2 range)
- Repetition 0
- Frequency 0.0
- Presence 0.00
SillyTavern
Prompt-post-processing: Strict or semi-strict