Local LLMs catch up: Qwen models now rival cloud AI

New Qwen3.6 and Qwen-Coder-Next models can run on consumer GPUs with performance matching cloud services like Claude and Gemini. The author reports usable speeds on dual RX6800 GPUs, with MoE variants offering fast inference. Meanwhile, free cloud tiers have degraded, making local models a viable alternative.