PRL-Bench: A Comprehensive Benchmark Evaluating LLMs' Capabilities in Frontier Physics Research

Hugging Face Daily Papers Papers

Summary

PRL-Bench is a comprehensive benchmark for evaluating LLMs' capabilities in frontier physics research, constructed from 100 curated Physical Review Letters papers across five physics subfields. The benchmark reveals significant gaps in current LLM performance (best scores below 50%), designed to test end-to-end research workflows, complex reasoning, and autonomous exploration.

The paradigm of agentic science requires AI systems to conduct robust reasoning and engage in long-horizon, autonomous exploration. However, current scientific benchmarks remain confined to domain knowledge comprehension and complex reasoning, failing to evaluate the exploratory nature and procedural complexity of real-world research. In this work, we present research-oriented evaluations in theoretical and computational physics, a natural testbed with comprehensive domain knowledge, complex reasoning, and verifiable end-to-end workflows without reliance on experiments. Here we introduce PRL-Bench (Physics Research by LLMs), a benchmark designed to systematically map the capability boundaries of LLMs in executing end-to-end physics research. Constructed from 100 curated papers from the latest issues of Physical Review Letters since August 2025 and validated by domain experts, PRL-Bench covers five major theory- and computation-intensive subfields of modern physics: astrophysics, condensed matter physics, high-energy physics, quantum information, and statistical physics. Each task in the benchmark is designed to replicate the core properties of authentic scientific research, including exploration-oriented formulation, long-horizon workflows, and objective verifiability, thereby reconstructing the essential reasoning processes and research workflows of real physics research. Evaluation across frontier models shows that performance remains limited, with the best overall score below 50, revealing a pronounced gap between current LLM capabilities and the demands of real scientific research. PRL-Bench serves a reliable testbed for accessing next generation AI scientists advancing AI systems toward autonomous scientific discovery.
Original Article
View Cached Full Text

Cached at: 04/20/26, 08:27 AM

Paper page - PRL-Bench: A Comprehensive Benchmark Evaluating LLMs’ Capabilities in Frontier Physics Research

Source: https://huggingface.co/papers/2604.15411 Authors:

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

Abstract

Current AI systems demonstrate limited capability in performing end-to-end physics research, highlighting a significant gap between existing language models and the demands of real scientific discovery.

The paradigm ofagentic science (https://huggingface.co/papers?q=agentic%20science)requires AI systems to conduct robust reasoning and engage in long-horizon,autonomous exploration (https://huggingface.co/papers?q=autonomous%20exploration). However, currentscientific benchmarks (https://huggingface.co/papers?q=scientific%20benchmarks)remain confined todomain knowledge (https://huggingface.co/papers?q=domain%20knowledge)comprehension and complex reasoning, failing to evaluate the exploratory nature and procedural complexity of real-world research. In this work, we present research-oriented evaluations in theoretical andcomputational physics (https://huggingface.co/papers?q=computational%20physics), a natural testbed with comprehensivedomain knowledge (https://huggingface.co/papers?q=domain%20knowledge), complex reasoning, and verifiableend-to-end workflows (https://huggingface.co/papers?q=end-to-end%20workflows)without reliance on experiments. Here we introducePRL-Bench (https://huggingface.co/papers?q=PRL-Bench)(Physics Research byLLMs (https://huggingface.co/papers?q=LLMs)), a benchmark designed to systematically map the capability boundaries ofLLMs (https://huggingface.co/papers?q=LLMs)in executing end-to-end physics research. Constructed from 100 curated papers from the latest issues of Physical Review Letters since August 2025 and validated by domain experts,PRL-Bench (https://huggingface.co/papers?q=PRL-Bench)covers five major theory- and computation-intensive subfields of modern physics: astrophysics, condensed matter physics, high-energy physics, quantum information, and statistical physics. Each task in the benchmark is designed to replicate the core properties of authenticscientific research (https://huggingface.co/papers?q=scientific%20research), including exploration-oriented formulation, long-horizon workflows, and objective verifiability, thereby reconstructing the essential reasoning processes and research workflows of real physics research. Evaluation across frontier models shows that performance remains limited, with the best overall score below 50, revealing a pronounced gap between current LLM capabilities and the demands of realscientific research (https://huggingface.co/papers?q=scientific%20research).PRL-Bench (https://huggingface.co/papers?q=PRL-Bench)serves a reliable testbed for accessing next generation AI scientists advancing AI systems toward autonomous scientific discovery.

View arXiv page (https://arxiv.org/abs/2604.15411)View PDF (https://arxiv.org/pdf/2604.15411)Add to collection (https://huggingface.co/login?next=%2Fpapers%2F2604.15411)

Get this paper in your agent:

hf papers read 2604\.15411

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2604.15411 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2604.15411 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2604.15411 in a Space README.md to link it from this page.

Collections including this paper1

Similar Articles

PRISM: A Multi-Dimensional Benchmark for Evaluating LLM Peer Reviewers

arXiv cs.CL

Introduces PRISM, a multi-dimensional benchmark for evaluating LLM-based peer reviewers across depth of analysis, novelty assessment, flaw identification, and constructiveness. Findings show LLMs match or beat humans on individual dimensions but lack balanced performance across all, suggesting they are best as supplements to human review.