Pith.

Computation and Language · arXiv:2604.01754 v2 · submitted 2 Apr 2026

LiveMathematicianBench: A Live Benchmark for Research-Level Mathematical Reasoning with Proof Sketches

Linyang He, Qiyao Yu, Hanze Dong, Baohao Liao, Xinxing Xu, Micah Goldblum, Jiang Bian, Nima Mesgarani

The paper says

“Overall, LiveMathematicianBench offers a scalable, contamination-resistant testbed for studying research-level mathematical reasoning in LLMs.”

from the abstract · picked out as the main claim by a classifier

Pith says

Pith has not reviewed this paper.

Get a Pith Review, $5

Reviews

Reviews by people

No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

Mathematical reasoning is a hallmark of human intelligence, and whether large language models (LLMs) can meaningfully perform it remains a central question in artificial intelligence and cognitive science. As LLMs are increasingly integrated into scientific workflows, rigorous evaluation of their mathematical capabilities becomes a practical necessity. Existing benchmarks are limited by synthetic settings and data contamination. We present LiveMathematicianBench, a dynamic multiple-choice benchmark for research-level mathematical reasoning built from recent arXiv papers published after model training cutoffs. By grounding evaluation in newly published theorems, it provides a realistic testbed beyond memorized patterns. The benchmark introduces a thirteen-category logical taxonomy of theorem types (e.g., implication, equivalence, existence, uniqueness), enabling fine-grained evaluation across reasoning forms. It employs a proof-sketch-guided distractor pipeline that uses high-level proof strategies to construct plausible but invalid answer choices reflecting misleading proof directions, increasing sensitivity to genuine understanding over surface-level matching. We also introduce a substitution-resistant mechanism to distinguish answer recognition from substantive reasoning. Evaluation shows the benchmark is far from saturated: Gemini-3.1-pro-preview, the best model, achieves only 43.5%. Under substitution-resistant evaluation, accuracy drops sharply: GPT-5.4 scores highest at 30.6%, while Gemini-3.1-pro-preview falls to 17.6%, below the 20% random baseline. A dual-mode protocol reveals that proof-sketch access yields consistent accuracy gains, suggesting models can leverage high-level proof strategies for reasoning. Overall, LiveMathematicianBench offers a scalable, contamination-resistant testbed for studying research-level mathematical reasoning in LLMs.

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-09-25 · no language-model review

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: claims first 0.11 · states limitation 0.24 · human participants 0.07 · controlled comparison 0.82 · randomized assignment 0.02 · states quantitative result 0.99 · compares with prior results 0.88 · mentions code or data availability 0.05

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v2). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
Mathematical reasoning is a hallmark of human intelligence, and whether large language models (LLMs) can meaningfully perform it remains a central question in artificial intelligence and cognitive science. background 1.00 0.03 0.02 asserted 0.95
As LLMs are increasingly integrated into scientific workflows, rigorous evaluation of their mathematical capabilities becomes a practical necessity. background 0.98 0.04 0.02 asserted 0.99
Existing benchmarks are limited by synthetic settings and data contamination. gap 0.95 0.04 0.02 asserted 1.00
We present LiveMathematicianBench, a dynamic multiple-choice benchmark for research-level mathematical reasoning built from recent arXiv papers published after model training cutoffs. contribution 0.98 0.93 0.08 asserted 1.00
By grounding evaluation in newly published theorems, it provides a realistic testbed beyond memorized patterns. contribution 0.69 0.56 0.08 asserted 0.99
The benchmark introduces a thirteen-category logical taxonomy of theorem types (e.g., implication, equivalence, existence, uniqueness), enabling fine-grained evaluation across reasoning forms. method 0.69 0.92 0.04 asserted 1.00
It employs a proof-sketch-guided distractor pipeline that uses high-level proof strategies to construct plausible but invalid answer choices reflecting misleading proof directions, increasing sensitivity to genuine understanding over surface-level matching. method 0.95 0.92 0.04 asserted 1.00
We also introduce a substitution-resistant mechanism to distinguish answer recognition from substantive reasoning. method 0.61 0.91 0.04 asserted 1.00
Evaluation shows the benchmark is far from saturated: Gemini-3.1-pro-preview, the best model, achieves only 43.5%. result 0.98 0.96 0.18 asserted 1.00
Under substitution-resistant evaluation, accuracy drops sharply: GPT-5.4 scores highest at 30.6%, while Gemini-3.1-pro-preview falls to 17.6%, below the 20% random baseline. result 1.00 0.95 0.17 asserted 1.00
A dual-mode protocol reveals that proof-sketch access yields consistent accuracy gains, suggesting models can leverage high-level proof strategies for reasoning. result 0.56 0.94 0.15 hedged possibility 0.90
Overall, LiveMathematicianBench offers a scalable, contamination-resistant testbed for studying research-level mathematical reasoning in LLMs. contribution 0.69 0.58 0.57 asserted 1.00

What kind of paper the classifier reports this is: dataset or benchmark 1.00 · evidence from observation or experiment 0.94

What was examined, and what was not

Sources read

  • arXiv listing abstract v2

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-25: 1 request, 8,880 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Cited by 7 Pith papers
How the literature uses it

13 arXiv papers cite this in 17 sentences

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

Papers citing it · 14 sentences without a confident label

Uses its data or values 3 papers
  • The comparison spans the official SearchQA and OfficeQA evaluation splits, SpreadsheetBench, and LiveMathematicianBench (abbreviated LiveMath) [ref] , with Qwen3.5-4B and GPT-5.5 as target models [ref] .

    Rethinking Self-Evolving Agents: Do We Still Need Prescribed Optimization Pipelines? · A frontier optimizer can improve skills without prescribed pipelines · 0.81

  • The main experiment reports each benchmark's native hard-success or exact-match score on the test partitions across SearchQA [ref] , SpreadsheetBench [ref] , DocVQA [ref] , LiveMath [ref] , and ALFWorld [ref] .

    SKILL-KD: Contrastive Skill Distillation for LLM Agents · Experiments · 0.89

  • Datasets We evaluate across five benchmarks spanning diverse domains: mathematical reasoning (LiveMathematicianBench (LiveMath) [ref] ), web search (SealQA [ref] ), spreadsheet manipulation (SpreadsheetBench (SpreadSheet) [ref] ), long-context document question answering OfficeQA [ref] ), and interactive embodied tasks (ALFWorld [ref] ).

    WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution · Experimental Setup · 0.81

What it cites · 5 sentences without a confident label

Cites as support 2 papers

Challenges and discussion

Know something about this paper? Leave a comment.

No comments yet.

Sign in to comment