128k context is not a limit of the model, that's a limit of implementation:
"Context Length: 262,144 natively and extensible up to 1,000,000 tokens."
https://huggingface.co/Qwen/Qwen3.8-27B