AI model builders mostly highlight unique benchmarks that act as flexible narrative tools for market positioning rather than standardized scientific measurements.
hub Baseline reference
MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark
Baseline reference. 55% of citing Pith papers use this work as a benchmark or comparison.
98
Pith papers citing it
13
external citations · Pith
Baseline
55% of classified citations
abstract
In the age of large-scale language models, benchmarks like the Massive Multitask Language Understanding (MMLU) have been pivotal in pushing the boundaries of what AI can achieve in language comprehension and reasoning across diverse domains. However, as models continue to improve, their performance on these benchmarks has begun to plateau, making it increasingly difficult to discern differences in model capabilities. This paper introduces MMLU-Pro, an enhanced dataset designed to extend the mostly knowledge-driven MMLU benchmark by integrating more challenging, reasoning-focused questions and expanding the choice set from four to ten options. Additionally, MMLU-Pro eliminates the trivial and noisy questions in MMLU. Our experimental results show that MMLU-Pro not only raises the challenge, causing a significant drop in accuracy by 16% to 33% compared to MMLU but also demonstrates greater stability under varying prompts. With 24 different prompt styles tested, the sensitivity of model scores to prompt variations decreased from 4-5% in MMLU to just 2% in MMLU-Pro. Additionally, we found that models utilizing Chain of Thought (CoT) reasoning achieved better performance on MMLU-Pro compared to direct answering, which is in stark contrast to the findings on the original MMLU, indicating that MMLU-Pro includes more complex reasoning questions. Our assessments confirm that MMLU-Pro is a more discriminative benchmark to better track progress in the field.
hub tools
citation-role summary
dataset 6
background 4
baseline 1
citation-polarity summary
claims ledger
- abstract In the age of large-scale language models, benchmarks like the Massive Multitask Language Understanding (MMLU) have been pivotal in pushing the boundaries of what AI can achieve in language comprehension and reasoning across diverse domains. However, as models continue to improve, their performance on these benchmarks has begun to plateau, making it increasingly difficult to discern differences in model capabilities. This paper introduces MMLU-Pro, an enhanced dataset designed to extend the mostly knowledge-driven MMLU benchmark by integrating more challenging, reasoning-focused questions and
co-cited works
representative citing papers
MMMU-Pro is a stricter multimodal benchmark that removes text-only solvable questions, augments options, and requires reading text from images, yielding substantially lower model scores of 16.8-26.9%.
LiveBench is a contamination-limited LLM benchmark with auto-scored challenging tasks from recent sources across math, coding, reasoning and more, where top models score below 70%.
Lynx partitions KV cache bits into anchor and residual streams for progressive transfer, enabling speculative decoding on partial data followed by verification to match BF16 accuracy at 4-bit-like TTFT.
A meta-benchmarking framework organizes 452 LLM benchmarks into 41 O*NET Generalized Work Activities and 38 BIAN domains, using discrimination-coverage-recency weights to scale K-factors in an Elo tournament for comparable financial-services scores.
ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite with largest gains at the 0.8B scale.
MemTrace shows that evidence utilization, not retrieval, is the dominant failure mode in LLM long-term memory systems across tested configurations.
Discourse-role labels on identical misleading context cause 56-84 percentage point shifts in LLMs adopting the injected wrong answer.
Geometry-aware FGW prior transfer plus PAC-Bayes residual adaptation reduces topology forgetting and raises average accuracy across continual multi-agent topology learning streams.
A new benchmark dataset drawn from Japan's National Assessment of Academic Ability supplies real exam layouts, diagrams, Japanese text, and nationwide student response distributions for evaluating multimodal LLMs.
TESSERA combines LLMs as local policy and evaluator with MCTS on knowledge graphs to compose mechanistic drug-disease explanations.
Star Elastic trains N nested submodels in a single post-training job on a parent reasoning LLM, supporting elastic budget control that matches or exceeds independent baselines while cutting training compute by up to 360x.
AOPD modifies on-policy distillation by using localized divergence minimization for non-positive advantages instead of negative reinforcement, yielding average gains of 4.09/8.34 over standard OPD on math reasoning benchmarks under strong/weak initialization.
Medical VLMs frequently select negated options that contradict visible chest X-ray findings, achieving only ~30% accuracy on direct presence probes, but a post-hoc consistency verifier raises accuracy above 95%.
MRI-Eval benchmark shows frontier LLMs scoring 93-97% on MRI MCQs but falling to 37-61% on stem-only questions, with GE scanner operations as the weakest category for all models.
Complex adversarial instructions induce positional collapse in LLMs, with extreme cases showing 99.9% concentration on a single response position and zero content sensitivity.
Evolutionary trees from LLM weights recover ground-truth training topologies and identify key datasets and layers through phenotypic analysis.
Distinct Leaf Enumeration (DLE) replaces stochastic self-consistency sampling with deterministic traversal of a truncated decoding tree to enumerate distinct leaves, increasing coverage and reducing redundant computation while improving performance on math, coding, and reasoning benchmarks.
A single 15B supernet checkpoint supports runtime switching between attention mixer placements for multiple decode speed presets while retaining 77-96% quality relative to the teacher model.
COVERT generates verifiable synthetic tool-use environments for RL by validated trajectory synthesis and oracle-preserving augmentations, improving tool-use accuracy on BFCL v3 and ACEBench while remaining complementary to SFT.
MARS fine-tunes autoregressive models to predict multiple tokens per step via continued training on instruction data, achieving 1.5-1.7x throughput while matching baseline accuracy and supporting real-time speed adjustment.
CapTrack shows post-training causes drift beyond facts, with instruction fine-tuning producing stronger behavioral changes than preference optimization across model families.
PRIMETIME generator reveals that LLM datetime parsing and arithmetic primitives are individually unreliable but fully learnable via fine-tuning, enabling frontier-level accuracy on event planning with small LoRA models.
Video-MMMU benchmark shows large multimodal models exhibit steep performance drops on higher cognitive tasks when learning from professional videos and lag significantly behind humans in knowledge acquisition.
citing papers explorer
-
Unsteady Metrics and Benchmarking Cultures of AI Model Builders
AI model builders mostly highlight unique benchmarks that act as flexible narrative tools for market positioning rather than standardized scientific measurements.
-
MMMU-Pro: A More Robust Multi-discipline Multimodal Understanding Benchmark
MMMU-Pro is a stricter multimodal benchmark that removes text-only solvable questions, augments options, and requires reading text from images, yielding substantially lower model scores of 16.8-26.9%.
-
LiveBench: A Challenging, Contamination-Limited LLM Benchmark
LiveBench is a contamination-limited LLM benchmark with auto-scored challenging tasks from recent sources across math, coding, reasoning and more, where top models score below 70%.
-
Lynx: Progressive Speculative Quantization for accelerating KV Transfer in Long-Context Inference
Lynx partitions KV cache bits into anchor and residual streams for progressive transfer, enabling speculative decoding on partial data followed by verification to match BF16 accuracy at 4-bit-like TTFT.
-
Meta-Benchmarks for Financial-Services LLM Evaluation
A meta-benchmarking framework organizes 452 LLM benchmarks into 41 O*NET Generalized Work Activities and 38 BIAN domains, using discrimination-coverage-recency weights to scale K-factors in an Elo tournament for comparable financial-services scores.
-
Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients
ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite with largest gains at the 0.8B scale.
-
MemTrace: Probing What Final Accuracy Misses in Long-Term Memory
MemTrace shows that evidence utilization, not retrieval, is the dominant failure mode in LLM long-term memory systems across tested configurations.
-
Discourse-Role Labels as Presentation-Time Variables for Context Use in Language Models
Discourse-role labels on identical misleading context cause 56-84 percentage point shifts in LLMs adopting the injected wrong answer.
-
MasFACT: Continual Multi-Agent Topology Learning via Geometry-Aware Posterior Transfer
Geometry-aware FGW prior transfer plus PAC-Bayes residual adaptation reduces topology forgetting and raises average accuracy across continual multi-agent topology learning streams.
-
Human-Grounded Multimodal Benchmark with 900K-Scale Aggregated Student Response Distributions from Japan's National Assessment of Academic Ability
A new benchmark dataset drawn from Japan's National Assessment of Academic Ability supplies real exam layouts, diagrams, Japanese text, and nationwide student response distributions for evaluating multimodal LLMs.
-
LLM-Guided Monte Carlo Tree Search over Knowledge Graphs: Composing Mechanistic Explanations for Drug-Disease Pairs
TESSERA combines LLMs as local policy and evaluator with MCTS on knowledge graphs to compose mechanistic drug-disease explanations.
-
Star Elastic: Many-in-One Reasoning LLMs with Efficient Budget Control
Star Elastic trains N nested submodels in a single post-training job on a parent reasoning LLM, supporting elastic budget control that matches or exceeds independent baselines while cutting training compute by up to 360x.
-
Asymmetric On-Policy Distillation: Bridging Exploitation and Imitation at the Token Level
AOPD modifies on-policy distillation by using localized divergence minimization for non-positive advantages instead of negative reinforcement, yielding average gains of 4.09/8.34 over standard OPD on math reasoning benchmarks under strong/weak initialization.
-
CXR-ContraBench: Benchmarking Negated-Option Attraction in Medical VLMs
Medical VLMs frequently select negated options that contradict visible chest X-ray findings, achieving only ~30% accuracy on direct presence probes, but a post-hoc consistency verifier raises accuracy above 95%.
-
MRI-Eval: A Tiered Benchmark for Evaluating LLM Performance on MRI Physics and GE Scanner Operations Knowledge
MRI-Eval benchmark shows frontier LLMs scoring 93-97% on MRI MCQs but falling to 37-61% on stem-only questions, with GE scanner operations as the weakest category for all models.
-
Instruction Complexity Induces Positional Collapse in Adversarial LLM Evaluation
Complex adversarial instructions induce positional collapse in LLMs, with extreme cases showing 99.9% concentration on a single response position and zero content sensitivity.
-
Analysis and Explainability of LLMs Via Evolutionary Methods
Evolutionary trees from LLM weights recover ground-truth training topologies and identify key datasets and layers through phenotypic analysis.
-
Efficient Test-Time Inference via Deterministic Exploration of Truncated Decoding Trees
Distinct Leaf Enumeration (DLE) replaces stochastic self-consistency sampling with deterministic traversal of a truncated decoding tree to enumerate distinct leaves, increasing coverage and reducing redundant computation while improving performance on math, coding, and reasoning benchmarks.
-
Super Apriel: One Checkpoint, Many Speeds
A single 15B supernet checkpoint supports runtime switching between attention mixer placements for multiple decode speed presets while retaining 77-96% quality relative to the teacher model.
-
Controllable and Verifiable Tool-Use Data Synthesis for Agentic Reinforcement Learning
COVERT generates verifiable synthetic tool-use environments for RL by validated trajectory synthesis and oracle-preserving augmentations, improving tool-use accuracy on BFCL v3 and ACEBench while remaining complementary to SFT.
-
MARS: Enabling Autoregressive Models Multi-Token Generation
MARS fine-tunes autoregressive models to predict multiple tokens per step via continued training on instruction data, achieving 1.5-1.7x throughput while matching baseline accuracy and supporting real-time speed adjustment.
-
CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training
CapTrack shows post-training causes drift beyond facts, with instruction fine-tuning producing stronger behavioral changes than preference optimization across model families.
-
PRIMETIME : Limits of LLMs in Temporal Primitives
PRIMETIME generator reveals that LLM datetime parsing and arithmetic primitives are individually unreliable but fully learnable via fine-tuning, enabling frontier-level accuracy on event planning with small LoRA models.
-
Video-MMMU: Evaluating Knowledge Acquisition from Multi-Discipline Professional Videos
Video-MMMU benchmark shows large multimodal models exhibit steep performance drops on higher cognitive tasks when learning from professional videos and lag significantly behind humans in knowledge acquisition.
-
VoiceBench: Benchmarking LLM-Based Voice Assistants
VoiceBench is the first benchmark for multi-faceted evaluation of LLM voice assistants using real and synthetic spoken instructions with speaker, environmental, and content variations.
-
Unified Audio Intelligence Without Regressing on Text Intelligence
Audex unifies audio understanding and generation on a strong text MoE backbone with multi-stage SFT plus text-only Cascade RL, matching open SOTA audio scores while mostly retaining text capability.
-
Addressing Over-Refusal in LLMs with Competing Rewards
SEAR trains one LLM via adversarial process rewards to explore harmful reasoning paths but flip to safe outputs, reducing over-refusal while preserving safety.
-
One-Step Gradient Delay is Not a Barrier for Large-Scale Asynchronous Pipeline Parallel LLM Pretraining
One-step gradient delay is optimizer-dependent rather than intrinsically unstable, with Muon and error-feedback correction enabling async pipeline parallelism to match synchronous performance on models up to 10B parameters.
-
SAKE: Software Architectural Knowledge Evaluation Benchmark for Large Language Models
Introduces SAKE benchmark with 2154 questions to assess LLMs on software architectural knowledge, showing high overall accuracy but marked gaps across categories.
-
To Reason or to Fabricate: Reasoning Without Shortcuts via Hint-Anchored Pairwise Aggregation
HIPPO is a new RL framework that uses hint-anchored pairwise aggregation to distinguish and promote authentic reasoning deduction in LLMs instead of shortcut memorization from data overlap.
-
Data and Evaluation Closed-Loop for Model Capability Enhancement
Proposes capability slices with dual taxonomies and mapping rules to form a closed loop converting benchmark failures into targeted data interventions, validated via two opposing case studies on BBH and math reasoning.
-
Pigeonholing: how bad prompts hurt models, causing collapse and mistakes
Unintentionally bad contexts (user suggestions or prior wrong assistant answers) cause LLMs to repeat errors, lose diversity, and flip stances, worsening with turns; RLVR on synthetic errors recovers 43–60% of the drop.
-
DRIFT: Refining Instruction Data via On-Policy Data Attribution
DRIFT applies on-policy influence functions with signed weighting and debiasing to attribute and refine SFT data, raising performance on 7B instruction and reasoning models over prior curation methods.
-
Beyond English benchmarks: clinical llm evaluation in Brazilian Portuguese
Creates the first bilingual clinical benchmark from Brazilian cases and reports that English performance advantage exists only in diagnosis retrieval, disappearing in the other three tasks.
-
LoMo: Local Modality Substitution for Deeper Vision-Language Fusion
LoMo is a lightweight data curation technique that locally substitutes text with images in prompts to enforce cross-modal invariance, yielding 2.67-2.82 point gains over standard SFT on two VLMs across 13 benchmarks.
-
The Routing Plateau: Understanding and Breaking the Accuracy Limits of LLM Routers
LLM routers across 21 methods on 5 benchmarks converge to similar accuracy below oracle due to learning global performance trends rather than fine-grained query signals.
-
Efficient Safety Benchmarking via Item Response Theory
Item Response Theory enables adaptive and fixed-subset item selection that reduces safety benchmark costs by 80-99.9% while preserving high correlation with full rankings.
-
Beyond Query Memorization: Large Language Model Routing with Query Decomposition and Historical Matching
DecoR routes LLM queries by decomposing them into capability dimensions and matching to historical examples, yielding higher accuracy and lower inference costs than direct-mapping routers on both in-distribution and OOD data.
-
Open-World Evaluations for Measuring Frontier AI Capabilities
Open-world evaluations using qualitative review of real-world tasks can give earlier warnings of frontier AI capabilities than automated benchmarks, as demonstrated by an AI agent publishing a simple iOS app with one minor human fix.
-
TeachArena: Are Language Agents Ready for Realistic Teaching Work?
A single three-surface benchmark shows frontier LLM agents pass ≈90% of bounded pedagogical-judgment tasks but only ≈36% of situated tutoring and ≈33% of LMS workflow tasks.
-
Rotation-Preserving Supervised Fine-Tuning
RPSFT improves the in-domain versus out-of-domain performance trade-off during LLM supervised fine-tuning by penalizing rotations in pretrained singular subspaces as a proxy for loss-sensitive directions.
-
An Interpretable and Scalable Framework for Evaluating Large Language Models
A majorization-minimization framework turns IRT into scalable matrix factorization subproblems for LLM evaluation, delivering orders-of-magnitude speedups with identifiability guarantees.
-
Option-Order Randomisation Reveals a Distributional Position Attractor in Prompted Sandbagging
Sandbagging prompts induce LLMs to adopt a low-entropy, content-invariant response-position attractor centered on E/F/G rather than deterministic tracking or random avoidance.
-
Large Language Models Decide Early and Explain Later
LLMs settle on their answer after a minority of CoT tokens and produce an average 760 more as post-decision explanation, enabling early stopping that saves 500 tokens per query at a 2% accuracy cost.
-
Decoupled DiLoCo for Resilient Distributed Pre-training
Decoupled DiLoCo enables asynchronous distributed pre-training with zero global downtime under simulated failures while preserving competitive performance on text and vision tasks.
-
COMPASS: COntinual Multilingual PEFT with Adaptive Semantic Sampling
COMPASS uses semantic clustering on multilingual embeddings to select auxiliary data for PEFT adapters, outperforming linguistic-similarity baselines on multilingual benchmarks while supporting continual adaptation.
-
Can LLMs Learn to Reason Robustly under Noisy Supervision?
Online Label Refinement lets LLMs learn robust reasoning from noisy supervision by correcting labels when majority answers show rising rollout success and stable history, delivering 3-4% gains on math and reasoning benchmarks even at high noise levels.
-
Structured Multi-Criteria Evaluation of Large Language Models with Fuzzy Analytic Hierarchy Process and DualJudge
Fuzzy AHP and DualJudge deliver more stable and calibrated LLM evaluations than direct scoring by breaking assessments into explicit criteria and adaptively fusing intuitive and deliberative judgments.
-
Rubrics to Tokens: Bridging Response-level Rubrics and Token-level Rewards in Instruction Following Tasks
RTT bridges response-level rubrics to token-level rewards via a relevance discriminator and intra-sample group normalization, yielding higher instruction and rubric accuracy than baselines.
-
When LLMs get significantly worse: A statistical approach to detect model degradations
A McNemar-based statistical test detects real degradations in optimized LLMs with controlled false positives, even for accuracy changes as small as 0.3%.