[julianlam]

Try llama.cpp and Qwen3.6-35B-A3B

Good balance of intelligence and speed.