Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MorphoBench: A Benchmark with Difficulty Adaptive to Model Reasoning

Xukai Wang, Xuanbo Liu, Mingrui Chen, Haitian Zhong, Xuanlin Yang, Bohan Zeng, Jinbo Hu, Hao Liang, Junbo Niu, Xuchen Li, Ruitao Wu, Ruichuan An, Yang Shi, Liu Liu, Xu-Yao Zhang, Qiang Liu, Zhouchen Lin, Wentao Zhang, Bin Dong

TL;DR

MorphoBench tackles the challenge of evaluating reasoning in large models by introducing a dynamically adaptive, multidisciplinary benchmark that modulates difficulty along two axes: understanding conditions and reasoning chains. It constructs a formal proof-graph framework to quantify problem difficulty, uses information-gap-based question modifications, and employs automatic generation with simulation to ensure validity; combined with category expansion, MorphoBench supports continual, domain-diverse evaluation. Empirical results across frontier models (e.g., o3, GPT-5, Gemini variants, Claude-4) show how difficulty adaptations reveal strengths and weaknesses in cross-domain reasoning and robustness to cues. The work provides a practical, extensible platform for diagnosing and improving model reasoning, with a public codebase enabling ongoing updates and broader benchmarking.

Abstract

With the advancement of powerful large-scale reasoning models, effectively evaluating the reasoning capabilities of these models has become increasingly important. However, existing benchmarks designed to assess the reasoning abilities of large models tend to be limited in scope and lack the flexibility to adapt their difficulty according to the evolving reasoning capacities of the models. To address this, we propose MorphoBench, a benchmark that incorporates multidisciplinary questions to evaluate the reasoning capabilities of large models and can adjust and update question difficulty based on the reasoning abilities of advanced models. Specifically, we curate the benchmark by selecting and collecting complex reasoning questions from existing benchmarks and sources such as Olympiad-level competitions. Additionally, MorphoBench adaptively modifies the analytical challenge of questions by leveraging key statements generated during the model's reasoning process. Furthermore, it includes questions generated using simulation software, enabling dynamic adjustment of benchmark difficulty with minimal resource consumption. We have gathered over 1,300 test questions and iteratively adjusted the difficulty of MorphoBench based on the reasoning capabilities of models such as o3 and GPT-5. MorphoBench enhances the comprehensiveness and validity of model reasoning evaluation, providing reliable guidance for improving both the reasoning abilities and scientific robustness of large models. The code has been released in https://github.com/OpenDCAI/MorphoBench.

MorphoBench: A Benchmark with Difficulty Adaptive to Model Reasoning

TL;DR

MorphoBench tackles the challenge of evaluating reasoning in large models by introducing a dynamically adaptive, multidisciplinary benchmark that modulates difficulty along two axes: understanding conditions and reasoning chains. It constructs a formal proof-graph framework to quantify problem difficulty, uses information-gap-based question modifications, and employs automatic generation with simulation to ensure validity; combined with category expansion, MorphoBench supports continual, domain-diverse evaluation. Empirical results across frontier models (e.g., o3, GPT-5, Gemini variants, Claude-4) show how difficulty adaptations reveal strengths and weaknesses in cross-domain reasoning and robustness to cues. The work provides a practical, extensible platform for diagnosing and improving model reasoning, with a public codebase enabling ongoing updates and broader benchmarking.

Abstract

With the advancement of powerful large-scale reasoning models, effectively evaluating the reasoning capabilities of these models has become increasingly important. However, existing benchmarks designed to assess the reasoning abilities of large models tend to be limited in scope and lack the flexibility to adapt their difficulty according to the evolving reasoning capacities of the models. To address this, we propose MorphoBench, a benchmark that incorporates multidisciplinary questions to evaluate the reasoning capabilities of large models and can adjust and update question difficulty based on the reasoning abilities of advanced models. Specifically, we curate the benchmark by selecting and collecting complex reasoning questions from existing benchmarks and sources such as Olympiad-level competitions. Additionally, MorphoBench adaptively modifies the analytical challenge of questions by leveraging key statements generated during the model's reasoning process. Furthermore, it includes questions generated using simulation software, enabling dynamic adjustment of benchmark difficulty with minimal resource consumption. We have gathered over 1,300 test questions and iteratively adjusted the difficulty of MorphoBench based on the reasoning capabilities of models such as o3 and GPT-5. MorphoBench enhances the comprehensiveness and validity of model reasoning evaluation, providing reliable guidance for improving both the reasoning abilities and scientific robustness of large models. The code has been released in https://github.com/OpenDCAI/MorphoBench.
Paper Structure (42 sections, 1 theorem, 13 equations, 12 figures, 4 tables)

This paper contains 42 sections, 1 theorem, 13 equations, 12 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Large Models
  4. Evaluation Benchmark for Large Models
  5. MorphoBench
  6. Data Collection
  7. Preliminary Analysis
  8. Reasoning as Path Search in a Proof Graph
  9. Question Modification and Information Gap
  10. Impact of Modifications on Reasoning Complexity
  11. Difficulty Adaptation
  12. Adaptation based on agent reasoning.
  13. Adaptation based on agent recognition.
  14. Adaptation for automatically generated questions.
  15. Category Expansion
  16. ...and 27 more sections

Key Result

Lemma 1

Let the original search graph be a single directed path which is the unique route from the start vertex $v_{0}$ to the goal vertex $v_{k}$. Embed an incompressible binary string $\tau$ of length $|\tau| = \Delta I$ bits into the graph by attaching $m$ dead-end (out-degree-one) edges while preserving $P$ as the only goal path. Then

Figures (12)

  • Figure 1: Overview of MorphoBench.
  • Figure 2: Testing examples from MorphoBench.
  • Figure 3: Demonstration of MorphoBench's problem difficulty adjustment pipelines.
  • Figure 4: Different large models' reasoning results on MorphoBench.
  • Figure 5: Diversity analysis of MorphoBench.
  • ...and 7 more figures

Theorems & Definitions (2)

  • Lemma 1
  • Proof 1