mlsanityData to 14 Oct 2026

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

arXiv 2509.18792 ↗First posted 23 Sep 2025cs.CLcited by 2

Not on the map. Why?

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.

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

11 of 30 references.

References · 11

  1. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  2. 2024Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preferencecited by 1,184
  3. 2024The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scalecited by 1,160
  4. 2024SimPO: Simple Preference Optimization with a Reference-Free Rewardcited by 963
  5. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 394
  6. 2017What do Neural Machine Translation Models Learn about Morphology?cited by 166
  7. 2025Are Sparse Autoencoders Useful? A Case Study in Sparse Probingcited by 115
  8. 2021Transformer visualization via dictionary learning: contextualized embedding as a linear superposition of transformer factorscited by 108
  9. 2025Fanar: An Arabic-Centric Multimodal Generative AI Platformcited by 67
  10. 2021How transfer learning impacts linguistic knowledge in deep NLP models?cited by 32
  11. 2023Can LLMs facilitate interpretation of pre-trained language models?cited by 16

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026There Is More to Refusal in Large Language Models than a Single Directioncited by 26
  2. 2026Med-SegLens: Latent-Level Model Diffing for Interpretable Medical Image Segmentationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Latent Performance Profiling of Large Language Models
  2. 2024Exploring the Trade-Offs: Quantization Methods, Task Difficulty, and Model Size in Large Language Models From Edge to Giant
  3. 2025Discovering Hierarchical Latent Capabilities of Language Models via Causal Representation Learning
  4. 2026An Interpretable and Scalable Framework for Evaluating Large Language Models
  5. 2025Towards Alignment-Centric Paradigm: A Survey of Instruction Tuning in Large Language Models
  6. 2026Benchmark Shadows: Data Alignment, Parameter Footprints, and Generalization in Large Language Models
  7. 2024Towards Uncovering How Large Language Model Works: An Explainability Perspective
  8. 2025Instruction Tuning with and without Context: Behavioral Shifts and Downstream Impact