_version_ 1866911735471931392
author Akhtar, Mubashara
Reuel, Anka
Soni, Prajna
Ahuja, Sanchit
Ammanamanchi, Pawan Sasanka
Rawal, Ruchit
Zouhar, Vilém
Yadav, Srishti
Whitehouse, Chenxi
Ki, Dayeon
Mickel, Jennifer
Choshen, Leshem
Šuppa, Marek
Batzner, Jan
Chim, Jenny
Sania, Jeba
Long, Yanan
Rahmani, Hossein A.
Knight, Christina
Nan, Yiyang
Raj, Jyoutir
Fan, Yu
Singh, Shubham
Sahoo, Subramanyam
Habba, Eliya
Gohar, Usman
Pawar, Siddhesh
Scholz, Robert
Subramonian, Arjun
Ni, Jingwei
Kochenderfer, Mykel
Koyejo, Sanmi
Sachan, Mrinmaya
Biderman, Stella
Talat, Zeerak
Ghosh, Avijit
Solaiman, Irene
author_facet Akhtar, Mubashara
Reuel, Anka
Soni, Prajna
Ahuja, Sanchit
Ammanamanchi, Pawan Sasanka
Rawal, Ruchit
Zouhar, Vilém
Yadav, Srishti
Whitehouse, Chenxi
Ki, Dayeon
Mickel, Jennifer
Choshen, Leshem
Šuppa, Marek
Batzner, Jan
Chim, Jenny
Sania, Jeba
Long, Yanan
Rahmani, Hossein A.
Knight, Christina
Nan, Yiyang
Raj, Jyoutir
Fan, Yu
Singh, Shubham
Sahoo, Subramanyam
Habba, Eliya
Gohar, Usman
Pawar, Siddhesh
Scholz, Robert
Subramonian, Arjun
Ni, Jingwei
Kochenderfer, Mykel
Koyejo, Sanmi
Sachan, Mrinmaya
Biderman, Stella
Talat, Zeerak
Ghosh, Avijit
Solaiman, Irene
contents Artificial intelligence benchmarks are an important mechanism for measuring model progress and guiding deployment decisions. However, benchmarks quickly "saturate", making it difficult to differentiate models and diminishing their long-term value. In this study, we define benchmark saturation and analyze it across 60 language model benchmarks using 14 properties that relate to saturation. We find that nearly half of the our benchmarks exhibit saturation, with rates increasing with age. Further, we find that resilience to saturation is impacted by expert-curation, not by public test data. Our results suggest that design choices can extend benchmark longevity and inform more durable evaluation approaches.
format Preprint
id arxiv_https___arxiv_org_abs_2602_16763
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle When AI Benchmarks Plateau: A Systematic Study of Benchmark Saturation
Akhtar, Mubashara
Reuel, Anka
Soni, Prajna
Ahuja, Sanchit
Ammanamanchi, Pawan Sasanka
Rawal, Ruchit
Zouhar, Vilém
Yadav, Srishti
Whitehouse, Chenxi
Ki, Dayeon
Mickel, Jennifer
Choshen, Leshem
Šuppa, Marek
Batzner, Jan
Chim, Jenny
Sania, Jeba
Long, Yanan
Rahmani, Hossein A.
Knight, Christina
Nan, Yiyang
Raj, Jyoutir
Fan, Yu
Singh, Shubham
Sahoo, Subramanyam
Habba, Eliya
Gohar, Usman
Pawar, Siddhesh
Scholz, Robert
Subramonian, Arjun
Ni, Jingwei
Kochenderfer, Mykel
Koyejo, Sanmi
Sachan, Mrinmaya
Biderman, Stella
Talat, Zeerak
Ghosh, Avijit
Solaiman, Irene
Artificial Intelligence
Artificial intelligence benchmarks are an important mechanism for measuring model progress and guiding deployment decisions. However, benchmarks quickly "saturate", making it difficult to differentiate models and diminishing their long-term value. In this study, we define benchmark saturation and analyze it across 60 language model benchmarks using 14 properties that relate to saturation. We find that nearly half of the our benchmarks exhibit saturation, with rates increasing with age. Further, we find that resilience to saturation is impacted by expert-curation, not by public test data. Our results suggest that design choices can extend benchmark longevity and inform more durable evaluation approaches.
title When AI Benchmarks Plateau: A Systematic Study of Benchmark Saturation
topic Artificial Intelligence
url https://arxiv.org/abs/2602.16763