Swift1.5-Qwen3.8-Flash-Next is phenomenal vs. base 3.8-Flash!

The Swift-1.5-Qwen3.8-Flash-Next model is highlighted as a significant improvement over the base 3.8-Flash, particularly for its ability to reduce excess reasoning. While statistically indistinguishable in overall quality based on McNemar exact p ≈ 0.29, it shows notable compression in C++ code, using only 29% of the base model's tokens compared to ~46% for Python/JavaScript. This makes it especially useful for C++ developers, despite taking 3 out of 6 losses in paired comparisons.