Microsoft released MAI-Code-1-Flash, a 137B parameter mixture of experts model with 5B active parameters that the company claims significantly outperforms Claude Haiku on the SWE-bench Pro coding benchmark. Community critics accuse the company of cherry-picking weak opponents for comparison, arguing the model merely matches the performance of much smaller and more efficient open-source alternatives like Qwen. In defense of the release, some users highlight Microsoft's emphasis on decontaminated training data and clean licensing, which could offer better real-world generalization and reduced legal liability.