PhySciBench benchmark shows current AI models achieve at most 33.5% accuracy on physical science tasks; DelveAgent framework improves accuracy by up to 7.5 points and cuts costs to one-third.
hub Canonical reference
Scienceagentbench: Toward rigorous assessment of language agents for data-driven scientific discovery
Canonical reference. 100% of citing Pith papers cite this work as background.
32
Pith papers citing it
Background
100% of classified citations
abstract
The advancements of large language models (LLMs) have piqued growing interest in developing LLM-based language agents to automate scientific discovery end-to-end, which has sparked both excitement and skepticism about their true capabilities. In this work, we call for rigorous assessment of agents on individual tasks in a scientific workflow before making bold claims on end-to-end automation. To this end, we present ScienceAgentBench, a new benchmark for evaluating language agents for data-driven scientific discovery. To ensure the scientific authenticity and real-world relevance of our benchmark, we extract 102 tasks from 44 peer-reviewed publications in four disciplines and engage nine subject matter experts to validate them. We unify the target output for every task to a self-contained Python program file and employ an array of evaluation metrics to examine the generated programs, execution results, and costs. Each task goes through multiple rounds of manual validation by annotators and subject matter experts to ensure its annotation quality and scientific plausibility. We also propose two effective strategies to mitigate data contamination concerns. Using ScienceAgentBench, we evaluate five open-weight and proprietary LLMs, each with three frameworks: direct prompting, OpenHands CodeAct, and self-debug. Given three attempts for each task, the best-performing agent can only solve 32.4% of the tasks independently and 34.3% with expert-provided knowledge. In addition, we evaluate OpenAI o1-preview with direct prompting and self-debug, which can boost the performance to 42.2%, demonstrating the effectiveness of increasing inference-time compute but with more than 10 times the cost of other LLMs. Still, our results underscore the limitations of current language agents in generating code for data-driven discovery, let alone end-to-end automation for scientific research.
hub tools
citation-role summary
background 5
dataset 2
citation-polarity summary
polarities
background 6representative citing papers
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.
MetaSyn is a stage-level benchmark of 442 meta-analyses showing LLM agents retrieve up to 90.9% of eligible studies but include at most 52.7% in their final reports.
Data2Story is a multi-agent framework that generates evidence-grounded multimodal articles from data, evaluated on 18 articles against human pieces for verifiability, angle coverage, and quality across human, rubric, and automated judges.
InquiTree shows LLM agents suffer from degrading critical capabilities during extended scientific interactions and perform worse on papers published after their training cutoffs.
Introduces the Matter to Mechanism benchmark of 2,645 structured instances and a composite metric suite for evaluating AI co-scientists on problem-to-hypothesis reasoning in battery materials research.
SCICONVBENCH is a new benchmark evaluating LLMs on multi-turn disambiguation and inconsistency resolution for task formulation in computational science, with frontier models reaching only 52.7% success on fluid mechanics disambiguation cases.
FML-Bench shows a simple greedy hill-climber nearly matches tree search on dense-opportunity tasks while an adaptive agent that broadens search on stagnation outperforms six baselines across 18 tasks.
A finite sheaf-theoretic framework ranks obstruction measures to identify when an AI agent's theory must deform within its language or extend to a new one, validated on a controlled transition benchmark.
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.
Agentic-imodels evolves scikit-learn regressors via an autoresearch loop to jointly boost predictive performance and LLM-simulatability, improving downstream agentic data science tasks by up to 73% on the BLADE benchmark.
LLM agents execute scientific tasks but fail to follow core scientific reasoning norms such as evidence consideration and belief revision based on refutations.
A survey of 1,250 papers organizes AI self-improvement along two axes—what is improved and loop closure—finding that demonstrated self-improvement strength tracks a verification hierarchy from formal verifiers down to intrinsic self-assessment.
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.
OPD-Evolver uses on-policy self-distillation in fast interaction and slow attribution loops to build agents with holistic memory competence, outperforming prior systems by up to 11.5% and allowing a 9B model to compete with much larger ones.
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.
SIGA is a coding-agent adapter using retrieval, procedural memory, and validation gates that raises success rate on GEOS from 0.720 to 0.789 while cutting variance 16x and matching expert quality in minutes instead of hours.
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.
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
LLM coding agents cannot reach the 10^{-4} relative accuracy required for gravitational wave modeling tasks and show systematic failures including metric misuse and result fabrication.
TREX automates the LLM training lifecycle via collaborative agents and tree-based exploration, delivering consistent performance gains across 10 real-world fine-tuning tasks in FT-Bench.
Gated KalmaNet uses exact Kalman gain computation with adaptive gating and Chebyshev iteration to improve SSM performance on long-context tasks over prior approximations like DeltaNet.
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 is a new benchmark showing that LLM coding agents fail to autonomously implement most realistic research extensions to prior AI papers.
citing papers explorer
-
Deep Research in Physical Sciences: A Multi-Agent Framework and Comprehensive Benchmark
PhySciBench benchmark shows current AI models achieve at most 33.5% accuracy on physical science tasks; DelveAgent framework improves accuracy by up to 7.5 points and cuts costs to one-third.
-
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.
-
MetaSyn: A Benchmark for LLM Agents on Meta-Analysis Articles from Nature Portfolio
MetaSyn is a stage-level benchmark of 442 meta-analyses showing LLM agents retrieve up to 90.9% of eligible studies but include at most 52.7% in their final reports.
-
Data Journalist Agent: Transforming Data into Verifiable Multimodal Stories
Data2Story is a multi-agent framework that generates evidence-grounded multimodal articles from data, evaluated on 18 articles against human pieces for verifiability, angle coverage, and quality across human, rubric, and automated judges.
-
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.
-
Matter to Mechanism: A Benchmark for AI Co-Scientists in Materials and Battery Research
Introduces the Matter to Mechanism benchmark of 2,645 structured instances and a composite metric suite for evaluating AI co-scientists on problem-to-hypothesis reasoning in battery materials research.
-
SCICONVBENCH: Benchmarking LLMs on Multi-Turn Clarification for Task Formulation in Computational Science
SCICONVBENCH is a new benchmark evaluating LLMs on multi-turn disambiguation and inconsistency resolution for task formulation in computational science, with frontier models reaching only 52.7% success on fluid mechanics disambiguation cases.
-
FML-bench: A Controlled Study of AI Research Agent Strategies from the Perspective of Search Dynamics
FML-Bench shows a simple greedy hill-climber nearly matches tree search on dense-opportunity tasks while an adaptive agent that broadens search on stagnation outperforms six baselines across 18 tasks.
-
Sheaf-Theoretic Transport and Obstruction for Detecting Scientific Theory Shift in AI Agents
A finite sheaf-theoretic framework ranks obstruction measures to identify when an AI agent's theory must deform within its language or extend to a new one, validated on a controlled transition benchmark.
-
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.
-
Agentic-imodels: Evolving agentic interpretability tools via autoresearch
Agentic-imodels evolves scikit-learn regressors via an autoresearch loop to jointly boost predictive performance and LLM-simulatability, improving downstream agentic data science tasks by up to 73% on the BLADE benchmark.
-
AI scientists produce results without reasoning scientifically
LLM agents execute scientific tasks but fail to follow core scientific reasoning norms such as evidence consideration and belief revision based on refutations.
-
Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loops
A survey of 1,250 papers organizes AI self-improvement along two axes—what is improved and loop closure—finding that demonstrated self-improvement strength tracks a verification hierarchy from formal verifiers down to intrinsic self-assessment.
-
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.
-
OPD-Evolver: Cultivating Holistic Agent Evolver via On-Policy Distillation
OPD-Evolver uses on-policy self-distillation in fast interaction and slow attribution loops to build agents with holistic memory competence, outperforming prior systems by up to 11.5% and allowing a 9B model to compete with much larger ones.
-
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.
-
Auto-Configuring Scientific Simulators with Lightweight Coding-Agent Adapters
SIGA is a coding-agent adapter using retrieval, procedural memory, and validation gates that raises success rate on GEOS from 0.720 to 0.789 while cutting variance 16x and matching expert quality in minutes instead of hours.
-
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.
-
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
-
gwBenchmarks: Stress-Testing LLM Agents on High-Precision Gravitational Wave Astronomy
LLM coding agents cannot reach the 10^{-4} relative accuracy required for gravitational wave modeling tasks and show systematic failures including metric misuse and result fabrication.
-
TREX: Automating LLM Fine-tuning via Agent-Driven Tree-based Exploration
TREX automates the LLM training lifecycle via collaborative agents and tree-based exploration, delivering consistent performance gains across 10 real-world fine-tuning tasks in FT-Bench.
-
Gated KalmaNet: A Fading Memory Layer Through Test-Time Ridge Regression
Gated KalmaNet uses exact Kalman gain computation with adaptive gating and Chebyshev iteration to improve SSM performance on long-context tasks over prior approximations like DeltaNet.
-
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.
-
DeepResearch Bench: A Comprehensive Benchmark for Deep Research Agents
DeepResearch Bench supplies 100 expert-crafted PhD-level tasks and two human-aligned evaluation frameworks to measure deep research agents on report quality and citation accuracy.
-
Marginal Advantage Accumulation for Memory-Driven Agent Self-Evolution
MAA formalizes alignability and comparability conditions and uses differential signals, EMA accumulation, and semantic identity merging to enable cross-batch operation-level evidence accumulation, outperforming batch-level baselines in 14 of 16 settings while matching online methods.
-
Experiment-as-Code Labs: A Declarative Stack for AI-Driven Scientific Discovery
The paper introduces Experiment-as-Code Labs as a declarative stack synthesizing AI agents, systems orchestration, and physical lab control for AI-driven discovery.
-
EvoMaster: A Foundational Evolving Agent Framework for Agentic Science at Scale
EvoMaster is an open-source agent framework whose iterative self-evolution loop achieves the highest average score on 9 of 10 scientific agent benchmarks compared to OpenHands, OpenClaw, and Codex.
-
A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence
The paper delivers the first systematic review of self-evolving agents, structured around what components evolve, when adaptation occurs, and how it is implemented.
-
From LLM Reasoning to Autonomous AI Agents: A Comprehensive Review
A survey consolidating benchmarks, agent frameworks, real-world applications, and protocols for LLM-based autonomous agents into a proposed taxonomy with recommendations for future research.
-
Beyond NL2Code: A Structured Survey of Multimodal Code Intelligence
A structured survey of multimodal code intelligence that formulates the field by code roles and organizes work into four domains while proposing verification-centered research directions.
-
Position: Multimodal Large Language Models Can Significantly Advance Scientific Reasoning
Position paper claims multimodal LLMs can significantly advance scientific reasoning and proposes a four-stage roadmap plus challenges and suggestions.