Darwin Family: MRI-Trust-Weighted Evolutionary Merging for Training-Free Scaling of Language-Model Reasoning

Benchmark Model Rank Results
multiple-choice-question-answering-mcqa-on-gpqa-diamondDarwin-35B-A3B-Opus–Accuracy: 90.0
reading-comprehension-on-raceDarwin-27B-Opus–Accuracy: 83.1
sentence-completion-on-hellaswagDarwin-27B-Opus–Accuracy: 87.0