The paper introduces Balance of Benchmarks (BoB), a framework that improves task-conditioned model comparison by weighting benchmark evidence based on semantic density, equating scores across varying difficulty levels, and pooling task-relevant residuals. BoB retains all eligible benchmark data while adjusting its influence, outperforming uniform averaging on the WildScores dataset with higher Spearman correlation, lower MAE, and better shortlist hit rates. The method also reduces ranking instability when benchmarks are repeated or paraphrased, and lowers retrospective regret in model selection.
By Jhen-Ke Lin, Hong-Yun Lin
The study evaluates zero‑shot essay scoring using sub‑3B open language models that run locally on a single 8 GB consumer GPU. Four instruction‑tuned models (Qwen2.5‑0.5B, 1.5B, 3B and SmolLM2‑1.7B) were tested on all eight ASAP‑AES prompts, comparing rubric‑decomposed versus holistic prompting, different aggregation methods, and trait‑mapping strategies. Results show rubric‑decomposed prompting consistently outperforms holistic prompting, trait‑mapping is sensitive to calibration, and longer essays reduce error, yet the best local configuration (macro QWK 0.388) still falls short of human agreement and a length‑only baseline, suggesting these models are best suited for formative, human‑supervised feedback.
By Nguyen Dung Son, Dang Quang Minh, Nguyen Huu Loi, Truong Viet Vu, Nguyen Thai Anh
arXiv:2609.09372v1 Announce Type: cross
Abstract: Although MMLU is widely adopted as a benchmark for calibrating general AI capabilities, we psychometrically demonstrate that its aggregate score prim...
By Dana Paquin, Riddhiman Jain
arXiv:2511. 04689v3 Announce Type: replace-cross Abstract: Evaluating large language models (LLMs) typically requires thousands of benchmark items, making the process expensive, slow, and increasingly impractical at scale.
By Peiyu Li, Xiuxiu Tang, Si Chen, Ying Cheng, Ronald Metoyer, Ting Hua, Nitesh V. Chawla
The paper demonstrates that Item Response Theory (IRT) can uncover meaningful structure in safety benchmarks for language models, allowing adaptive item selection to approximate full benchmark rankings with Spearman’s ρ > 0.90 while cutting evaluation costs by at least 80% and up to 99.9% on some suites. It also proposes a static method to extract a small, informative subset of items that can be reused across models, achieving 80–99.8% cost savings. These findings show that psychometric techniques can make safety evaluation more efficient without sacrificing ranking accuracy.
By Fabio Spagliardi, M\'irian Silva, Ayan Datta, Aiden Zhou, Vamshi Bonagiri, Diogo Cruz
The paper investigates how benchmark contamination—leakage of test items into training data—affects large language model (LLM) leaderboards. By comparing original test items with semantically equivalent paraphrases, the authors measure contamination as a violation of anchor-item invariance and find that it inflates absolute scores but rarely changes model rankings. Across 47 public models and 74 finetuned models on four benchmarks, the rank correlation between standard and paraphrase-controlled leaderboards is 0.997, with only a handful of cases showing differential contamination that could alter rankings.
By Xingyao Xiao (Stanford University), Yihong Cheng (City University of Macau)
The study analyzes 373,019 judgments from LLM‑scored benchmarks, decomposing variance into system, item, judge, and interaction components via generalizability theory. It finds that with a single judge, generalizability converges to a ceiling determined by the system‑by‑judge variance, which is substantially lower in pairwise preference settings, allowing one judge to suffice. The research also reveals significant biases in presentation order and highlights that many published win‑rate claims fall below the measured floor of the benchmarks.
By Atul Anand
arXiv:2608. 05086v1 Announce Type: new Abstract: Language models differ in how safely they behave and these differences are measured by safety benchmarks.
By Joshua Fonseca Rivera (Independent), Neil Shah (Independent), David Demitri Africa (UK AI Security Institute), Konstantinos Voudouris (UK AI Security Institute)
arXiv:2608.21382v1 Announce Type: new
Abstract: Multiple-choice benchmarks fix the questions and the correct answers, but not the harness: the order of the options, the wording of the prompt, and whe...
By V. S. Raghu Parupudi
The paper argues that verbalized confidence—once viewed as overconfident and coarse—has become the preferred soft‑scoring method for LLM‑as‑a‑Judge on top‑tier proprietary models released after 2025. Experiments on SummEval, AggreFact, and HelpSteer2 across up to 18 LLMs show that log‑probabilities are no longer the best signal, and that adding an overconfidence advisory and self‑debate further improves calibration and robustness. The authors note that these enhancements incur little accuracy loss on post‑2025 models but do affect pre‑2025 ones, highlighting a compatibility shift in how confidence should be measured.
By Yu-Chung Hsiao
Judge-specific sensitivity is useful for aggregating pairwise LLM evaluations, but its interpretation depends on which systematic presentation effects the ranking model includes. We introduce OSCAR, a...
The study investigates whether small differences in leaderboard rankings between large language models (LLMs) are robust to changes in benchmark composition. Using item‑level responses from five benchmarks and a spectral approximation to multidimensional item‑response theory, the authors find that while overall rankings remain highly correlated, a significant portion (30.9–47.1%) of near‑tie pairs reverse order when benchmark items are recomposed based on low differential item functioning. This suggests that sub‑one‑percentage‑point leaderboard gaps may not reliably reflect true model superiority.
By Qiaoyuan Zheng, Yiqu Yang