KCSAT-ML is a new benchmark built from 664 Korean College Scholastic Ability Test mathematics problems, including a 339‑item core set with official per‑item error rates from nationwide cohorts of hundreds of thousands of examinees. The benchmark introduces the Difficulty‑aligned Reasoning Gain (DRG) metric, which evaluates whether a model’s mistakes align with items humans find hard or easy, revealing distinct patterns in how vision‑language and large language models perform across difficulty levels. The dataset and code are publicly available at https://github.com/naver-ai/KCSAT-ML.
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...
arXiv:2511. 21692v3 Announce Type: replace-cross Abstract: We investigate how well large language models (LLMs) generalize across different task difficulties, a key question for effective data curation and evaluation.
The paper investigates how aggregate benchmark scores can obscure item-level changes when commercial large language model APIs are upgraded. By querying 900 benchmark items across three GPT-5.4 to GPT-5.6 upgrades, the authors classify each item as reliably improved, reliably regressed, practically equivalent, or inconclusive, revealing that both improvements and regressions coexist within the same upgrade. The study shows that even large aggregate gains can hide up to 8.3% of reliably regressed items, and that strict versus loose scoring can dramatically alter perceived performance changes.
arXiv:2608. 05643v1 Announce Type: new Abstract: Test-time scaling improves LLM reasoning by using additional inference compute, but wider sampling alone can suffer from diminishing returns: new rollouts often repeat existing answer patterns instead of adding useful reasoning diversity.
The paper introduces the Grounded Integration Measure (GIM), a benchmark of 820 expert‑authored problems designed to test models on tasks that integrate multiple cognitive operations such as constraint satisfaction, state tracking, epistemic vigilance, and audience calibration. GIM emphasizes realistic, broadly accessible knowledge rather than specialized expertise, and uses a judge‑aware 2‑parameter logistic IRT model to produce robust ability estimates across 53 model‑thinking‑level configurations. The authors provide a comprehensive leaderboard of 22 models and 47 test configurations, and conduct an extensive study on how test‑time compute affects model capability, finding that configuration choices like thinking budget and quantization can be as influential as model selection itself. whyItMatters":"By focusing on integration of multiple cognitive domains, GIM offers a more realistic assessment of model reasoning capabilities than benchmarks that either overemphasize memorization or abstract reasoning alone."
arXiv:2608. 09351v1 Announce Type: cross Abstract: Test-time scaling improves LLM accuracy but multiplies inference cost, making the accuracy gained per unit of compute the metric that matters in deployment.
arXiv:2604. 11996v2 Announce Type: replace-cross Abstract: Should we trust Large Language Models (LLMs) with high accuracy?
arXiv:2607. 10139v1 Announce Type: cross Abstract: Selecting the correct answer from a pool of candidate reasoning chains is the engine of test-time scaling, yet the standard selectors each carry a cost: self-consistency inherits the errors of the single model it resamples, and trained reward models need labeled data and transfer poorly off-distribution.
arXiv:2602. 20629v3 Announce Type: replace Abstract: As Large Language Models (LLMs) saturate elementary benchmarks, the research frontier has shifted from generation to the reliability of automated evaluation.
arXiv:2509. 14704v3 Announce Type: replace Abstract: Benchmark saturation and training-data contamination increasingly obscure whether reported gains in large language models (LLMs) reflect genuine advances in reasoning or familiarity with recurring patterns in benchmark problems.
The paper investigates item-sensitivity—whether a language model’s choice depends on the specific input—in a forced-choice signalling task derived from the board game Deception: Murder in Hong Kong. Across seven models, two families, a post‑training ablation, and three scoring rules, every tested cell shows item‑sensitivity, yet many are statistically indistinguishable from random choice and some perform worse than random. The authors term this phenomenon "consistency without alignment" and argue it undermines evaluations that rely solely on item‑sensitivity, permutation consistency, or self‑consistency without an independent reference.
arXiv:2606. 01682v1 Announce Type: cross Abstract: Selecting the best response from multiple small-model samples using a stronger scorer is a simple inference-time strategy, but fails when the small model has already committed to incorrect reasoning paths.