#2736 Why Qwen3.6-35B-A3B run more and more slowly than Qwen3.8-Flash-Next

本文

I updated to 0.1.44 and used this cmd to build strata engine to run Qwen3.6-35B-A3B 
```python
START-HERE.bat --yes --family qwen36 --model UD-Q4_K_XL --no-start --build --gguf-dir D:\ollama\models\Qwen3.6-35B-A3B-UD-Q4_K_XL --data-dir D:\code5\Strata-data
```
I compared Qwen3.8-Flash-Next_IQ2_XS and Qwen3.6-35B-A3B-UD-Q4_K_XL with 128k context.
My Windows 11 computer has 12GB of VRAM and 64GB of RAM.
The output speed is:
```python
Qwen3.8-Flash-Next_IQ2_XS  70tok/s
Qwen3.6-35B-A3B-UD-Q4_K_XL 40tok/s
```
Why? Is this normal?
The computational complexity of these two models should be similar, right?

<img width="1296" height="217" alt="Image" src="https://github.com/user-attachments/assets/f6a8e7ee-a2a9-4efa-8718-b71b206fdde0" />

But why is there so much difference in speed?