# Beyond the Leaderboard: Understanding Performance Disparities in Large Language Models via Model Diffing

arXiv: https://arxiv.org/abs/2509.18792 · HTML: https://mlsanity.com/paper/2509.18792/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 11 of its 30 references that are in the map.
- First posted on arXiv 23 Sep 2025.

## Builds on (most cited first, 11 of 11)

- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference](https://mlsanity.com/paper/2403.04132/) (2024)
- [The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale](https://mlsanity.com/paper/2406.17557/) (2024)
- [SimPO: Simple Preference Optimization with a Reference-Free Reward](https://mlsanity.com/paper/2405.14734/) (2024)
- [Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2](https://mlsanity.com/paper/2408.05147/) (2024)
- [What do Neural Machine Translation Models Learn about Morphology?](https://mlsanity.com/paper/1704.03471/) (2017)
- [Are Sparse Autoencoders Useful? A Case Study in Sparse Probing](https://mlsanity.com/paper/2502.16681/) (2025)
- [Transformer visualization via dictionary learning: contextualized embedding as a linear superposition of transformer factors](https://mlsanity.com/paper/2103.15949/) (2021)
- [Fanar: An Arabic-Centric Multimodal Generative AI Platform](https://mlsanity.com/paper/2501.13944/) (2025)
- [How transfer learning impacts linguistic knowledge in deep NLP models?](https://mlsanity.com/paper/2105.15179/) (2021)
- [Can LLMs facilitate interpretation of pre-trained language models?](https://mlsanity.com/paper/2305.13386/) (2023)

## Built on by (most cited first, 2 of 2)

- [There Is More to Refusal in Large Language Models than a Single Direction](https://mlsanity.com/paper/2602.02132/) (2026)
- [Med-SegLens: Latent-Level Model Diffing for Interpretable Medical Image Segmentation](https://mlsanity.com/paper/2602.10508/) (2026)

## Abstract (arXiv)

As fine-tuning becomes the dominant paradigm for improving large language models (LLMs), understanding what changes during this process is increasingly important. Traditional benchmarking often fails to explain why one model outperforms another. In this work, we use model diffing, a mechanistic interpretability approach, to analyze the specific capability differences between Gemma-2-9b-it and a SimPO-enhanced variant. Using crosscoders, we identify and categorize latent representations that differentiate the two models. We find that SimPO acquired latent concepts predominantly enhance safety mechanisms (+32.8%), multilingual capabilities (+43.8%), and instruction-following (+151.7%), while its additional training also reduces emphasis on model self-reference (-44.1%) and hallucination management (-68.5%). Our analysis shows that model diffing can yield fine-grained insights beyond leaderboard metrics, attributing performance gaps to concrete mechanistic capabilities. This approach offers a transparent and targeted framework for comparing LLMs.