local model usage has 8x'd in @cline over the last 4 months invest in qwen3 coder