Six months ago, local models were not competitive, but now Qwen-Next 3.8 and 3.8 27b outperform Sonnet 5.5 low and medium, demonstrating rapid progress in edge AI. The author highlights Qwen-Next 3.8's reliability in complex tasks, citing a case where it corrected a project that GPT-Sol-6-High failed. This marks a pivotal shift toward cutting-edge local model performance.