This dude casually runs Qwen3.5-27B fully local on Telegram.
No APIs, no costs, no data sent anywhere, all sandboxed with NVIDIA NemoClaw on a DGX Spark.
9 t/s and smooth chats. The future might be ...
If it's going through Telegram it's sent somewhere. And what is the price of the local machine to run the model ? I don't get it.
1
