# Benchmark Illusion: Disagreement among LLMs and Its Scientific Consequences

arXiv: https://arxiv.org/abs/2602.11898 · HTML: https://mlsanity.com/paper/2602.11898/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 5 of its 19 references that are in the map.
- First posted on arXiv 12 Feb 2026.

## Builds on (most cited first, 5 of 5)

- [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023)
- [Underspecification Presents Challenges for Credibility in Modern Machine Learning](https://mlsanity.com/paper/2011.03395/) (2020)
- [Testing the Reliability of ChatGPT for Text Annotation and Classification: A Cautionary Remark](https://mlsanity.com/paper/2304.11085/) (2023)
- [Large Language Model Hacking: Quantifying the Hidden Risks of Using LLMs for Text Annotation](https://mlsanity.com/paper/2509.08825/) (2025)
- [Scientific production in the era of Large Language Models](https://mlsanity.com/paper/2601.13187/) (2026)

## Built on by (most cited first, 4 of 4)

- [MemDelta: Controlled Baselines and Hidden Confounds in Agent Memory Evaluation](https://mlsanity.com/paper/2606.29914/) (2026)
- [DFAH-Bench: Benchmarking Observable Agent Instability in Financial Decision-Making](https://mlsanity.com/paper/2607.20491/) (2026)
- [Benchmark Radar: A Living Database and Search Engine for AI Benchmarks and Evaluation](https://mlsanity.com/paper/2609.11115/) (2026)
- [A latent dimension of Condorcet's jury theorem for multiple AI advisers](https://mlsanity.com/paper/2609.14438/) (2026)

## Abstract (arXiv)

Benchmarks underpin how progress in large language models (LLMs) is measured and trusted. Yet our analyses reveal that apparent convergence in benchmark accuracy can conceal deep epistemic divergence. Using two major reasoning benchmarks - MMLU-Pro and GPQA - we show that LLMs achieving comparable accuracy still disagree on 16-66% of items, and 16-38% among top-performing frontier models. These discrepancies suggest distinct error profiles for different LLMs. When such models are used for scientific data annotation and inference, their hidden disagreements propagate into research results: in re-analyses of published studies in education and political science, switching the annotation model can change estimated treatment effects by more than 80%, and in some cases reverses their sign. Together, these findings illustrate a benchmark illusion, where equal accuracy may conceal disagreement, with model choice becoming a hidden yet consequential variable for scientific reproducibility.