Although the benchmark results are quite different from reality, I felt that Qwen3.6 27b's coding ability was pretty good for a local model, and the actual benchmark results are also quite impressive. Among open models, it surpasses the performance of existing large models despite its size of 27b.
