ReplicatorBench evaluates LLM agents on replicating social and behavioral science claims across retrieval, computation, and interpretation stages, finding strength in experiment execution but weakness in resource retrieval.
hub Mixed citations
PaperBench: Evaluating AI's Ability to Replicate AI Research
Mixed citation behavior. Most common role is background (67%).
56
Pith papers citing it
24
external citations · Pith
Background
67% of classified citations
abstract
We introduce PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research. Agents must replicate 20 ICML 2024 Spotlight and Oral papers from scratch, including understanding paper contributions, developing a codebase, and successfully executing experiments. For objective evaluation, we develop rubrics that hierarchically decompose each replication task into smaller sub-tasks with clear grading criteria. In total, PaperBench contains 8,316 individually gradable tasks. Rubrics are co-developed with the author(s) of each ICML paper for accuracy and realism. To enable scalable evaluation, we also develop an LLM-based judge to automatically grade replication attempts against rubrics, and assess our judge's performance by creating a separate benchmark for judges. We evaluate several frontier models on PaperBench, finding that the best-performing tested agent, Claude 3.5 Sonnet (New) with open-source scaffolding, achieves an average replication score of 21.0%. Finally, we recruit top ML PhDs to attempt a subset of PaperBench, finding that models do not yet outperform the human baseline. We open-source our code (https://github.com/openai/preparedness) to facilitate future research in understanding the AI engineering capabilities of AI agents.
hub tools
citation-role summary
background 6
dataset 2
method 1
citation-polarity summary
representative citing papers
Frontier coding agents surpass published Nature-family SOTA on only 17.8% of 90 sealed scientific tasks, mostly by recasting problems as supervised ML rather than inventing methods.
PixJail automates construction of paper-specific attack modules and unified evaluation pipelines for text-to-image jailbreaks, reproducing eleven methods with 2.1% average and 0% median error.
Only two of ten advanced AI agents finish a 500-day simulated CEO challenge above the starting cash, and none surpass a hand-tuned rule-based baseline.
SafeClawBench supplies 600 staged adversarial tasks and three separate endpoints that show semantic acceptance, audit evidence, and sandbox-observed harm are distinct failure modes in tool-using LLM agents.
InquiTree shows LLM agents suffer from degrading critical capabilities during extended scientific interactions and perform worse on papers published after their training cutoffs.
TensorBench is a new benchmark of 199 tasks on a tensor framework used to evaluate seven coding agents, yielding pass rates from 22.1% to 64.8% with low inter-agent agreement.
DeployBench is a new benchmark of 51 research-artifact deployment tasks where four LLMs with OpenHands achieve 7.8-51% pass rates, with failures mostly from agents stopping after weaker self-checks than the paper requires.
AutoMedBench evaluates AI agents on long-horizon medical workflows across five stages and finds validation and submission as dominant failure points based on thousands of runs.
WebGameBench is a new benchmark that evaluates coding agents on building browser-native games from frozen specifications, with runtime browser evaluation showing best agents reach 76.9% usable rate but only 20.2% excellent rate.
Collider-Bench is a new benchmark showing that current LLM agents cannot reliably reproduce LHC analyses at the level of a physicist-in-the-loop.
AI agents handle individual data-loading and reformatting steps on neuroscience datasets but rarely complete fully error-free end-to-end pipelines, and AI judges are unreliable without ground-truth references.
AI CFD Scientist autonomously discovers a Spalart-Allmaras runtime correction reducing lower-wall Cf RMSE by 7.89% on the periodic hill at Reh=5600 while using a vision-language gate to detect 14 of 16 silent failures missed by solver checks.
AcademiClaw is a new benchmark of 80 student-sourced academic tasks where the best frontier AI agents achieve only a 55% pass rate.
A purpose-built, staged LLM agent correctly sets up and executes software analysis tools on 33 of 35 benchmark tasks, outperforming general-purpose agent baselines by at least 17 percentage points.
FactReview extracts claims from ML papers, positions them via literature retrieval, and verifies them through code execution, labeling each as Supported, Partially supported, or In conflict, as shown in a CompGCN case study.
New Text-to-Big SQL metrics show that LLM agents must balance accuracy with cost and speed at scale, where GPT-4o trades some accuracy for up to 12x speedup and GPT-5.2 proves more cost-effective than Gemini 3 Pro on large inputs.
Evalet applies functional fragmentation to deliver fragment-level qualitative analysis of LLM evaluations, with a user study showing 48% more misalignment detections than holistic scoring.
Pre-execution scope gating works only when the judge sees the user's request, with GLM-5.2 achieving F1=0.66 against an expert reference of 0.78.
Paper-replication makes coding agents complete SciML paper replications only when every recorded claim has provenance, comparison evidence, and report coverage in a validated workspace.
An empirical study evaluating tool-augmented LLM agents on 243 real-world energy analytics problems across data retrieval, knowledge interpretation, and quantitative modeling using domain-specific tools and multi-dimensional scoring.
Arbor combines a coordinator, executors, and a hypothesis tree to enable cumulative autonomous research, outperforming Codex and Claude Code by over 2.5x on six real tasks and reaching 86.36% Any Medal on MLE-Bench Lite.
Skill coverage measures which natural-language skill constraints an LLM agent trajectory exercises and passes, revealing low coverage on SkillsBench and enabling a 16% recovery of failed tasks via targeted skill emphasis.
Proposes agentic framework-based reproduction with a slot-binding interface to turn 16 PHM papers into standardized, assumption-aware benchmark implementations.
citing papers explorer
-
ReplicatorBench: Benchmarking LLM Agents for Replicability in Social and Behavioral Sciences
ReplicatorBench evaluates LLM agents on replicating social and behavioral science claims across retrieval, computation, and interpretation stages, finding strength in experiment execution but weakness in resource retrieval.
-
NatureBench: Can Coding Agents Match the Published SOTA of Nature-Family Papers?
Frontier coding agents surpass published Nature-family SOTA on only 17.8% of 90 sealed scientific tasks, mostly by recasting problems as supervised ML rather than inventing methods.
-
PixJail: Self-Evolving Paper-to-Pipeline Reproduction for Text-to-Image Jailbreak Evaluation
PixJail automates construction of paper-specific attack modules and unified evaluation pipelines for text-to-image jailbreaks, reproducing eleven methods with 2.1% average and 0% median error.
-
CEO-Bench: Can Agents Play the Long Game?
Only two of ten advanced AI agents finish a 500-day simulated CEO challenge above the starting cash, and none surpass a hand-tuned rule-based baseline.
-
SafeClawBench: Separating Semantic, Audit-Evidence, and Sandbox Harm in Tool-Using LLM Agents
SafeClawBench supplies 600 staged adversarial tasks and three separate endpoints that show semantic acceptance, audit evidence, and sandbox-observed harm are distinct failure modes in tool-using LLM agents.
-
InquiTree: Evaluating AI Agents in the Scientific Inquiry Loop with Paper-Derived Research Trees
InquiTree shows LLM agents suffer from degrading critical capabilities during extended scientific interactions and perform worse on papers published after their training cutoffs.
-
TensorBench: Benchmarking Coding Agents on a Compiler-Based Tensor Framework
TensorBench is a new benchmark of 199 tasks on a tensor framework used to evaluate seven coding agents, yielding pass rates from 22.1% to 64.8% with low inter-agent agreement.
-
DeployBench: Benchmarking LLM Agents for Research Artifact Deployment
DeployBench is a new benchmark of 51 research-artifact deployment tasks where four LLMs with OpenHands achieve 7.8-51% pass rates, with failures mostly from agents stopping after weaker self-checks than the paper requires.
-
AutoMedBench: Towards Medical AutoResearch with Agentic AI Models
AutoMedBench evaluates AI agents on long-horizon medical workflows across five stages and finds validation and submission as dominant failure points based on thousands of runs.
-
WebGameBench: Requirement-to-Application Evaluation for Coding Agents via Browser-Native Games
WebGameBench is a new benchmark that evaluates coding agents on building browser-native games from frozen specifications, with runtime browser evaluation showing best agents reach 76.9% usable rate but only 20.2% excellent rate.
-
Collider-Bench: Benchmarking AI Agents with Particle Physics Analysis Reproduction
Collider-Bench is a new benchmark showing that current LLM agents cannot reliably reproduce LHC analyses at the level of a physicist-in-the-loop.
-
Neurodata Without Boredom: Benchmarking Agentic AI for Data Reuse
AI agents handle individual data-loading and reformatting steps on neuroscience datasets but rarely complete fully error-free end-to-end pipelines, and AI judges are unreliable without ground-truth references.
-
AI CFD Scientist: Toward Open-Ended Computational Fluid Dynamics Discovery with Physics-Aware AI Agents
AI CFD Scientist autonomously discovers a Spalart-Allmaras runtime correction reducing lower-wall Cf RMSE by 7.89% on the periodic hill at Reh=5600 while using a vision-language gate to detect 14 of 16 silent failures missed by solver checks.
-
AcademiClaw: When Students Set Challenges for AI Agents
AcademiClaw is a new benchmark of 80 student-sourced academic tasks where the best frontier AI agents achieve only a 55% pass rate.
-
Evaluating LLM Agents on Automated Software Analysis Tasks
A purpose-built, staged LLM agent correctly sets up and executes software analysis tools on 33 of 35 benchmark tasks, outperforming general-purpose agent baselines by at least 17 percentage points.
-
FactReview: Evidence-Grounded Reviews with Literature Positioning and Execution-Based Claim Verification
FactReview extracts claims from ML papers, positions them via literature retrieval, and verifies them through code execution, labeling each as Supported, Partially supported, or In conflict, as shown in a CompGCN case study.
-
Both Ends Count! Just How Good are LLM Agents at "Text-to-Big SQL"?
New Text-to-Big SQL metrics show that LLM agents must balance accuracy with cost and speed at scale, where GPT-4o trades some accuracy for up to 12x speedup and GPT-5.2 proves more cost-effective than Gemini 3 Pro on large inputs.
-
Evalet: Evaluating Large Language Models through Functional Fragmentation
Evalet applies functional fragmentation to deliver fragment-level qualitative analysis of LLM evaluations, with a user study showing 48% more misalignment detections than holistic scoring.
-
ScopeJudge: Cost-Aware Pre-Execution Gating for Offensive Security Agents
Pre-execution scope gating works only when the judge sees the user's request, with GLM-5.2 achieving F1=0.66 against an expert reference of 0.78.
-
Coding-agents can replicate scientific machine learning papers
Paper-replication makes coding agents complete SciML paper replications only when every recorded claim has provenance, comparison evidence, and report coverage in a validated workspace.
-
How Do Tool-Augmented LLM Agents Perform on Real-World Energy Analytics Tasks?
An empirical study evaluating tool-augmented LLM agents on 243 real-world energy analytics problems across data retrieval, knowledge interpretation, and quantitative modeling using domain-specific tools and multi-dimensional scoring.
-
Toward Generalist Autonomous Research via Hypothesis-Tree Refinement
Arbor combines a coordinator, executors, and a hypothesis tree to enable cumulative autonomous research, outperforming Codex and Claude Code by over 2.5x on six real tasks and reaching 86.36% Any Medal on MLE-Bench Lite.
-
Skill Coverage: A Test Adequacy Metric for Agent Skills
Skill coverage measures which natural-language skill constraints an LLM agent trajectory exercises and passes, revealing low coverage on SkillsBench and enabling a 16% recovery of failed tasks via targeted skill emphasis.
-
From paper to benchmark: agentic, framework-based reproduction of under-specified methods in machine health intelligence
Proposes agentic framework-based reproduction with a slot-binding interface to turn 16 PHM papers into standardized, assumption-aware benchmark implementations.
-
Benchmarks are Not Enough: RAMP for Runtime Assessing of Agentic Models in Production Systems
RAMP evaluates 15 models on production-like serial workflows and reports completion rates collapsing from 100% to 20% with none finishing the full pipeline and costs varying by three orders of magnitude.
-
ScientistOne: Towards Human-Level Autonomous Research via Chain-of-Evidence
ScientistOne introduces Chain-of-Evidence and an audit system that achieves zero hallucinated references, perfect score verification, and top method-code alignment while matching or beating human experts on five frontier tasks and generalizing to six more.
-
How Far Are We From True Auto-Research?
ResearchArena shows that agent-generated papers fail top-tier acceptance standards primarily due to fabricated results, underpowered experiments, and plan-execution mismatches that vary sharply by agent.
-
ArtifactLinker: Linking Scientific Artifacts for Automatic State-of-the-Art Discovery
ArtifactLinker frames SOTA discovery as missing-link prediction on an artifact graph of models and datasets, with a two-stage ranking-plus-verification pipeline and a new benchmark of 14k artifacts.
-
MLReplicate: Benchmarking Autonomous Research Systems for Machine Learning Reproducibility
MLReplicate benchmark evaluates six autonomous systems on 45 manuscripts from ICML 2025 papers, finding that automated reviews accept flawed outputs with fabricated claims while human review exposes methodological failures, and that the cheapest system outperforms the most expensive by a wide margin
-
Agentic Discovery of Neural Architectures: AIRA-Compose and AIRA-Design
Multi-agent LLM systems discover new Transformer and hybrid architectures that outperform Llama 3.2 at 1B scale and approach human SOTA on long-range benchmarks.
-
ARA: Agentic Reproducibility Assessment For Scalable Support Of Scientific Peer-Review
ARA uses LLMs to build workflow graphs linking sources, methods, and outputs in papers, then scores reproducibility, reaching ~61% accuracy on 213 ReScience C articles and outperforming priors on ReproBench and GoldStandardDB.
-
Evaluation-driven Scaling for Scientific Discovery
SimpleTES scales test-time evaluation in LLMs to discover state-of-the-art solutions on 21 scientific problems across six domains, outperforming frontier models and optimization pipelines with examples like 2x faster LASSO and new Erdos constructions.
-
Frontier-Eng: Benchmarking Self-Evolving Agents on Real-World Engineering Tasks with Generative Optimization
Frontier-Eng is a new benchmark for generative optimization in engineering where agents iteratively improve designs under fixed interaction budgets using executable verifiers, with top models like GPT 5.4 showing limited success.
-
Self-Preference Bias in Rubric-Based Evaluation of Large Language Models
Self-preference bias persists in rubric-based LLM evaluation even with fully objective, programmatically verifiable rubrics, and can shift subjective medical-chat scores by up to ~10 points.
-
In-Place Test-Time Training
In-Place TTT adapts LLM MLP projection matrices at test time with a next-token-aligned objective and chunk-wise updates, enabling better long-context performance as a drop-in enhancement.
-
AutoSOTA: An End-to-End Automated Research System for State-of-the-Art AI Model Discovery
A eight-agent system automatically reproduces and improves SOTA models from top AI papers, reporting 105 surpassing variants on a curated 125-paper set.
-
Towards Verifiable and Self-Correcting AI Physicists for Quantum Many-Body Simulations
QMP-Bench supplies a realistic test set for AI on quantum many-body problems while PhysVEC uses integrated verifiers to turn unreliable LLM generations into code that passes both syntax and physics checks, outperforming baselines.
-
Automating Computational Reproducibility in Social Science: Comparing Prompt-Based and Agent-Based Approaches
Agent-based AI workflows repair injected reproducibility failures in R social-science code at 69-96% success, substantially outperforming prompt-based LLM approaches at 31-79%.
-
CFDLLMBench: A Benchmark Suite for Evaluating Large Language Models in Computational Fluid Dynamics
CFDLLMBench is a new benchmark suite with CFDQuery, CFDCodeBench, and FoamBench to evaluate LLMs on graduate-level CFD knowledge, numerical reasoning, and context-dependent code implementation.
-
RExBench: Can coding agents autonomously implement AI research extensions?
RExBench is a new benchmark showing that LLM coding agents fail to autonomously implement most realistic research extensions to prior AI papers.
-
ResearchClawBench: A Benchmark for End-to-End Autonomous Scientific Research
A 40-task, 10-domain benchmark finds current auto-research agents and LLMs score only ~20–26 on re-discovering real published scientific artifacts.
-
Clarus: Coordinating Autonomous Research Agents toward Web-Scale Scientific Collaboration
Clarus is a four-layer collaboration infrastructure with a project-agent-resource model that reformulates research as an open, traceable, multi-participant process.
-
Agon: An Autonomous Large-Scale Omnidisciplinary Research System Built on Prompt Economy
Agon is a new autonomous research system using prompt economy loops across 444 iterations to demonstrate scalable omnidisciplinary research and a taxonomy separating machine-fixable failures from those needing human judgment.
-
Review the Code, Not the Story: A Vision and Protocol for Code-First Peer Review
Proposes a code-first peer review protocol using AI infrastructure to execute research artifacts and generate claim-evidence review packages for human reviewers.
-
Sibyl-AutoResearch: Autonomous Research Needs Self-Evolving Trial-and-Error Harnesses, Not Paper Generators
Sibyl-AutoResearch introduces self-evolving trial-and-error harnesses with auditable conversion units that link trial signals to updated research behaviors and harness repairs in autonomous systems.
-
ReproScore: Separating Readiness from Outcome in Research Software Reproducibility Assessment
ReproScore separates readiness (26 static sub-metrics) from outcome (execution probes) and shows near-zero correlation between them on 423 repositories, validating the separation.
-
RESCORE: LLM-Driven Simulation Recovery in Control Systems Research Papers
RESCORE recovers task-coherent simulations from 40.7% of 500 CDC papers via a three-component LLM agent pipeline and claims a 10X speedup over manual human replication.
-
Kimi K2.5: Visual Agentic Intelligence
Kimi K2.5 combines joint text-vision training with an Agent Swarm parallel orchestration framework to reach claimed state-of-the-art results on coding, vision, reasoning, and agent tasks while cutting latency up to 4.5 times.
-
CodeWiki: Evaluating AI's Ability to Generate Holistic Documentation for Large-Scale Codebases
CodeWiki presents a unified framework for repository-level documentation across seven languages using hierarchical decomposition, recursive multi-agent processing, and multi-modal synthesis, outperforming DeepWiki by 4.73% on CodeWikiBench.
-
Kimi K2: Open Agentic Intelligence
Kimi K2 is a 1-trillion-parameter MoE model that leads open-source non-thinking models on agentic benchmarks including 65.8 on SWE-Bench Verified and 66.1 on Tau2-Bench.