Soohak is a 439-problem mathematician-curated benchmark where frontier LLMs reach at most 30.4% on research math challenges and no model exceeds 50% on refusal for ill-posed problems.
hub Canonical reference
FrontierMath: A Benchmark for Evaluating Advanced Mathematical Reasoning in AI
Canonical reference. 70% of citing Pith papers cite this work as background.
36
Pith papers citing it
Background
70% of classified citations
abstract
We introduce FrontierMath, a benchmark of hundreds of original, exceptionally challenging mathematics problems crafted and vetted by expert mathematicians. The questions cover most major branches of modern mathematics -- from computationally intensive problems in number theory and real analysis to abstract questions in algebraic geometry and category theory. Solving a typical problem requires multiple hours of effort from a researcher in the relevant branch of mathematics, and for the upper end questions, multiple days. FrontierMath uses new, unpublished problems and automated verification to reliably evaluate models while minimizing risk of data contamination. Current state-of-the-art AI models solve under 2% of problems, revealing a vast gap between AI capabilities and the prowess of the mathematical community. As AI systems advance toward expert-level mathematical abilities, FrontierMath offers a rigorous testbed that quantifies their progress.
hub tools
citation-role summary
background 7
dataset 3
citation-polarity summary
representative citing papers
A Lean-verified multi-agent system produces a catalogue of 14,116 quantum codes with transversal diagonal gates for small parameters, extracts infinite families, and resolves specific distance-3 cases with constructions and no-go proofs.
CritPt benchmark shows state-of-the-art LLMs reach only 5.7% average accuracy on full-scale unpublished physics research tasks, rising to about 10% with coding tools.
LeanMarathon uses four contract-scoped agents on an evolving blueprint coordinated by a two-stage orchestrator to formalize seven theorems from Erdős problems in Lean, proving 258 lemmas with no sorry across three runs.
CrowdMath is a new dataset of annotated collaborative math proof discussions where frontier LLMs achieve 83-88% on next-post prediction but only 0.42 macro-F1 on identifying contribution roles.
GTBench is a new curriculum-grounded benchmark showing GPT-5 performs strongly on basic graph theory tasks but all models, including it, struggle more on advanced proofs with notable evaluator disagreements.
A new benchmark of 9,415 Lean 4 specifications derived from 2,772 scraped Python property-based tests, plus a three-agent LLM transpilation pipeline and proof-generation baselines.
The paper introduces a multi-turn interactive benchmark using 474 executable games to evaluate LLMs on evidence acquisition, belief updating, contextual robustness, and metacognitive adaptation, revealing large performance gaps and sensitivity to perturbations.
Introduces the CUSP benchmark across 4760 events and finds frontier AI models can pick plausible directions but fail to predict whether or when scientific advances will occur, with performance varying by domain and insensitive to training cutoffs.
Formal Conjectures is a Lean 4 benchmark containing 2615 formalized problems with 1029 open conjectures, designed to evaluate automated mathematical reasoning and proof discovery.
Self-play between LLMs for problem authoring and solving, scored via Rasch modeling, shows that authoring and solving skills are partially decoupled and that the benchmark difficulty evolves with new models.
Small 7B reasoning models were fine-tuned on synthetic and curated QFT problems using RL and SFT, yielding performance gains, error analysis, and public release of data and traces.
A harness for AI agents enabled construction of a Rust library with 100+ problem types and 200+ reduction rules for NP-hard problems in three months.
k-server-bench formulates potential-function discovery for the k-server conjecture as a code-based inequality-satisfaction task; current agents fully solve the resolved k=3 case and reduce violations on the open k=4 case.
DEONTICBENCH is a new benchmark of 6,232 deontic reasoning tasks from U.S. legal domains where frontier LLMs reach only ~45% accuracy and symbolic Prolog assistance plus RL training still fail to solve tasks reliably.
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.
LRMs show a large production-evaluation gap on the VAIR dataset with valid answers but invalid reasoning, driven by answer confirmation bias as evidenced by CoT analysis, linear probes, and causal patching.
Larger models succeed on rare and complex tasks by reducing gradient interference from common tasks, allowing rare-task features to accumulate, as shown via synthetic task mixtures and OLMo pretraining from 4M to 4B parameters.
Bidirectional Evolutionary Search augments autoregressive expansion with evolutionary recombination operators and dense backward subgoal feedback to produce better candidates than standard best-of-N or tree search for language model self-improvement.
Lean-GAP is a dataset of 430 graduate algebra problems formalized in Lean 4 from Dummit and Foote, with a described pipeline for autoformalization and verification plus analysis of challenges.
RMA, a multi-agent system with structured memory and iterative feedback loops, solves 8 out of 10 research-level math problems on the new First Proof benchmark and outperforms GPT-5.2R and Aletheia according to expert evaluation.
OPT-BENCH trains LLMs on NP-hard optimization via quality-aware RLVR, achieving 93.1% success rate and 46.6% quality ratio on Qwen2.5-7B while outperforming GPT-4o and transferring gains to other domains.
An empirical evaluation of 22 agentic frameworks on BBH, GSM8K, and ARC benchmarks shows stable performance in 12 frameworks but highlights orchestration failures and weaker mathematical reasoning.
XpertBench provides 1,346 rubric-scored expert tasks showing leading LLMs achieve a maximum ~66% success rate and ~55% mean score across domains.
citing papers explorer
-
AI for Mathematics: Progress, Challenges, and Prospects
AI for math combines task-specific architectures and general foundation models to support research and advance AI reasoning capabilities.