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

ACL ARR 2026 May Submission2611 Authors

Keywords: model merging, evolutionary optimization, large language models, reasoning, transfer learning

Abstract:

We present Darwin Family, a framework for training-free evolutionary merging of large language models via gradient-free weight-space recombination. We ask whether frontier-level reasoning performance can be improved without additional training by reorganizing latent capabilities already encoded in existing checkpoints.
Darwin introduces three key ideas: (i) a 14-dimensional adaptive merge genome enabling fine-grained component and block-level recombination; (ii) MRITrust Fusion, which adaptively balances diagnostic layer-importance signals with evolutionary search through a learnable trust parameter; and (iii) an Architecture Mapper that enables cross-architecture breeding between heterogeneous model families.
Empirically, the flagship Darwin-27BOpus achieves 86.9% on GPQA Diamond, ranking #6 among 1,252 evaluated models, and outperforming its fully trained foundation model without any gradient-based training. Across scales from 4B to 35B parameters, Darwin models consistently improve over their parents, support recursive multi-generation evolution, and enable a training-free evolutionary
merge that combines Transformer and Mamba-based components. Together, the Darwin Family demonstrates that diagnostic-guided evolutionary merging is a practical and reproducible alternative to costly post-training pipelines for reasoning-centric language models.

Paper Type: Long

Research Area: Efficient Methods for NLP

Research Area Keywords: model merging, evolutionary optimization, large language models, reasoning, transfer learning, model combination, neural architecture optimization

Contribution Types: NLP engineering experiment, Approaches to low-compute settings (efficiency), Publicly available software and/or pre-trained models

Languages Studied: English, Korean

EMNLP 2026 AI Reviewing Experiment: no

Submission Number: 2611