Heterophily-informed Message Passing

Haishan Wang, Arno Solin, Vikas Garg

Abstract

Graph neural networks (GNNs) are known to be vulnerable to oversmoothing due to their implicit homophily assumption. We mitigate this problem with a novel scheme that regulates the aggregation of messages, modulating the type and extent of message passing locally thereby preserving both the low and high-frequency components of information. Our approach relies solely on learnt embeddings, obviating the need for auxiliary labels, thus extending the benefits of heterophily-aware embeddings to broader applications,, generative modelling. Our experiments, conducted across various data sets and GNN architectures, demonstrate performance enhancements and reveal heterophily patterns across standard classification benchmarks. Furthermore, application to molecular generation showcases notable performance improvements on chemoinformatics benchmarks.

Benchmark Results

Tested on 11 datasets 34 baseline values reported too low Non-standard metric on 2 datasets

Tables show presented results in global context (best 3 papers + GCN + MLP).

Classic

PubMed

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	LGTL GT global	0.9518 ±0.0011	Adaptive Tokenization: On the Hop-Overpriority Problem in Tokenized Graph Learning Models
2	MixHop (LO) code ↗ GNN global	0.9513 ±0.0038	LOGIN: A Large Language Model Consulted Graph Neural Network Training Framework code ↗
3	GraphSAGE (LO) code ↗ GNN global	0.9511 ±0.0036	LOGIN: A Large Language Model Consulted Graph Neural Network Training Framework code ↗
18	GCN GNN baseline actual #18 vs paper #130 112 places worse	0.9296 ±0.0015 best reported	actual #18 vs paper #130 112 places worse paper 0.8850 ±0.0050 best at pub: Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning (ICLR 2024) best today: Adaptive Tokenization: On the Hop-Overpriority Problem in Tokenized Graph Learning Models (2025) paper 0.8850 ±0.0050 Δ at pub 0.0154 Δ today 0.0446
20	GAT GNN baseline actual #20 vs paper #200 180 places worse	0.9233 ±0.0018 best reported	actual #20 vs paper #200 180 places worse paper 0.8710 ±0.0050 best at pub: Tackling Over-Smoothing for General Graph Convolutional Networks (2020) best today: Adaptive Tokenization: On the Hop-Overpriority Problem in Tokenized Graph Learning Models (2025) paper 0.8710 ±0.0050 Δ at pub 0.0260 Δ today 0.0523
95	GraphSAGE+mix. Hyb baseline	0.8930 ±0.0040	This paper
114	GraphSAGE+hom. GNN baseline	0.8890 ±0.0040	This paper
116	SAGE GNN global	0.8881 ±0.0015	Efficient Text-Attributed Graph Learning through Selective Annotation and Graph Alignment
122	GCN+hom. GNN baseline	0.8870 ±0.0050	This paper
125	GraphSAGE GNN baseline actual #125 vs paper #156 31 places worse	0.8864 ±0.0030 best reported	actual #125 vs paper #156 31 places worse paper 0.8810 ±0.0050 best at pub: Non-backtracking Graph Neural Networks (TMLR 2023) paper 0.8810 ±0.0050 Δ at pub 0.0054
127	GCN+mix. Hyb baseline	0.8860 ±0.0050	This paper
155	GAT+hom. GNN baseline	0.8820 ±0.0050	This paper
157	GIN GNN baseline actual #157 vs paper #228 71 places worse	0.8810 best reported	actual #157 vs paper #228 71 places worse paper 0.8610 ±0.0080 best at pub: UGSL: A Unified Framework for Benchmarking Graph Structure Learning (2023) paper 0.8610 ±0.0080 Δ at pub 0.0200
167	GAT+mix. Hyb baseline	0.8780 ±0.0070	This paper
190	GIN+hom. GNN baseline	0.8740 ±0.0080	This paper
210	MLP code ↗ Trad global	0.8690 ±0.0074	On the Equivalence of Graph Convolution and Mixup code ↗
222	GIN+mix. Hyb baseline	0.8660 ±0.0090	This paper
282	GCN+het. GNN baseline	0.8390 ±0.0050	This paper
368	GIN+het. GNN baseline	0.8110 ±0.1490	This paper
496	GAT+het. GNN baseline	0.7960 ±0.0090	This paper

Amazon-Photo

Global leaderboard →

Non-default metric. This paper reports ROC-AUC; the dataset default is Accuracy. Values below use ROC-AUC only.

Sorted by ROC-AUC. Highlighted rows are from other papers.

#	Model	ROC-AUC ↑	Paper
1	GraphSAGE GNN baseline	0.9550 ±0.0040	This paper
2	GraphSAGE+hom. GNN baseline	0.9550 ±0.0040	This paper
3	GraphSAGE+mix. Hyb baseline	0.9520 ±0.0050	This paper
4	GraphSAGE+het. GNN baseline	0.9500 ±0.0060	This paper
5	GCN+mix. Hyb baseline	0.9460 ±0.0060	This paper
6	GCN GNN baseline	0.9440 ±0.0050	This paper
7	GCN+hom. GNN baseline	0.9440 ±0.0040	This paper
8	GAT GNN baseline	0.9410 ±0.0050	This paper
9	GAT+mix. Hyb baseline	0.9410 ±0.0030	This paper
10	GAT+hom. GNN baseline	0.9360 ±0.0060	This paper
11	GAT+het. GNN baseline	0.8730 ±0.0130	This paper
12	GCN+het. GNN baseline	0.8540 ±0.0160	This paper
13	GIN+mix. Hyb baseline	0.7820 ±0.1360	This paper
14	GIN+het. GNN baseline	0.3840 ±0.1570	This paper
15	GIN+hom. GNN baseline	0.3770 ±0.0800	This paper
16	GIN GNN baseline	0.3530 ±0.0710	This paper

Amazon-Computers

Global leaderboard →

Non-default metric. This paper reports ROC-AUC; the dataset default is Accuracy. Values below use ROC-AUC only.

Sorted by ROC-AUC. Highlighted rows are from other papers.

#	Model	ROC-AUC ↑	Paper
1	GCN GNN baseline	0.9160 ±0.0050	This paper
2	GCN+hom. GNN baseline	0.9150 ±0.0050	This paper
3	GCN+mix. Hyb baseline	0.9120 ±0.0060	This paper
4	GAT GNN baseline	0.9120 ±0.0030	This paper
5	GAT+mix. Hyb baseline	0.9110 ±0.0030	This paper
6	GAT+hom. GNN baseline	0.9090 ±0.0030	This paper
7	GraphSAGE GNN baseline	0.9090 ±0.0030	This paper
8	GraphSAGE+hom. GNN baseline	0.9060 ±0.0040	This paper
9	GraphSAGE+mix. Hyb baseline	0.9030 ±0.0050	This paper
10	GraphSAGE+het. GNN baseline	0.8940 ±0.0060	This paper
11	GAT+het. GNN baseline	0.8400 ±0.0200	This paper
12	GCN+het. GNN baseline	0.7830 ±0.0140	This paper
13	GIN+mix. Hyb baseline	0.5730 ±0.1470	This paper
14	GIN+het. GNN baseline	0.4310 ±0.0930	This paper
15	GIN+hom. GNN baseline	0.4280 ±0.0380	This paper
16	GIN GNN baseline	0.4120 ±0.0310	This paper

Heterophilic Graphs

Texas

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	P^2GNN Hyb global	0.9672 ±0.0147	$P^2$GNN: Two Prototype Sets to boost GNN Performance
2	DoG code ↗ ext data GNN global	0.9660 ±0.0210	Diffusion on Graph: Augmentation of Graph Structure for Node Classification code ↗
3	GPF-plus code ↗ ext data GNN global	0.9583 ±0.0419	ProG: A Graph Prompt Learning Benchmark code ↗
34	MLP Trad global	0.9230 ±0.0070	Design your own universe: a physics-informed agnostic method for enhancing graph neural networks
53	GraphSAGE GNN baseline actual #53 vs paper #261 208 places worse	0.9000 ±0.0365 best reported	actual #53 vs paper #261 208 places worse paper 0.7900 ±0.0120 best at pub: A Quasi-Wasserstein Loss for Learning Graph Neural Networks (WWW 2023) best today: Graph Neural Networks Need Cluster-Normalize-Activate Modules (NeurIPS 2024) paper 0.7900 ±0.0120 Δ at pub 0.1100 Δ today 0.1100
189	SAGE GNN global	0.8269	Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning
211	GAT GNN baseline actual #211 vs paper #544 333 places worse	0.8162 ±0.0645 best reported	actual #211 vs paper #544 333 places worse paper 0.5780 ±0.0460 best at pub: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) best today: Equivariance Everywhere All At Once: A Recipe for Graph Foundation Models (2025) paper 0.5780 ±0.0460 Δ at pub 0.2302 Δ today 0.2382
249	GraphSAGE+mix. Hyb baseline	0.7970 ±0.0590	This paper
257	GCN GNN baseline actual #257 vs paper #532 275 places worse	0.7933 ±0.0447 best reported	actual #257 vs paper #532 275 places worse paper 0.5840 ±0.0460 best at pub: Graph Wave Networks (WWW 2025) paper 0.5840 ±0.0460 Δ at pub 0.2093
303	GraphSAGE+het. GNN baseline	0.7590 ±0.0680	This paper
326	GIN GNN baseline actual #326 vs paper #550 224 places worse	0.7460 ±0.0080 best reported	actual #326 vs paper #550 224 places worse paper 0.5700 ±0.0690 best at pub: Unifying over-smoothing and over-squashing in graph neural networks: A physics informed approach and beyond (2023) best today: Design your own universe: a physics-informed agnostic method for enhancing graph neural networks (2024) paper 0.5700 ±0.0690 Δ at pub 0.1760 Δ today 0.1760
384	GraphSAGE+hom. GNN baseline	0.7050 ±0.0550	This paper
501	GCN+het. GNN baseline	0.6000 ±0.0770	This paper
509	GAT+het. GNN baseline	0.5970 ±0.0550	This paper
511	GAT+hom. GNN baseline	0.5950 ±0.0490	This paper
512	GIN+het. GNN baseline	0.5950 ±0.0800	This paper
525	GAT+mix. Hyb baseline	0.5920 ±0.0470	This paper
531	GCN+hom. GNN baseline	0.5860 ±0.0510	This paper
533	GIN+mix. Hyb baseline	0.5840 ±0.0680	This paper
538	GIN+hom. GNN baseline	0.5810 ±0.0680	This paper
554	GCN+mix. Hyb baseline	0.5680 ±0.0530	This paper

Cornell

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	P^2GNN Hyb global	0.9541 ±0.0272	$P^2$GNN: Two Prototype Sets to boost GNN Performance
2	ACM-GCN GNN baseline actual #2 vs paper #2	0.9480 ±0.0380	actual #2 vs paper #2 paper 0.9480 ±0.0380 prev best: Simplifying Graph Convolutional Networks with Redundancy-Free Neighbors (2025) paper 0.9480 ±0.0380 Δ at pub 0.0909
3	RF-GCN GNN global	0.9429	Simplifying Graph Convolutional Networks with Redundancy-Free Neighbors
22	GPRGNN GNN baseline actual #22 vs paper #22	0.9157 ±0.0196 best reported	actual #22 vs paper #22 paper 0.9140 ±0.0070 best at pub: Simplified PCNet with Robustness (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.9140 ±0.0070 Δ at pub 0.0017
24	MLP Trad global	0.9136 ±0.0070	Node-oriented Spectral Filtering for Graph Neural Networks
68	H2GCN GNN baseline actual #68 vs paper #71 3 places worse	0.8620 ±0.0470	actual #68 vs paper #71 3 places worse paper 0.8620 ±0.0470 prev best: Design your own universe: a physics-informed agnostic method for enhancing graph neural networks (2024) Improvement within reported uncertainty; not highlighted. paper 0.8620 ±0.0470 Δ at pub 0.0110
84	ACMP GNN baseline	0.8540 ±0.0700	This paper
138	SAGE GNN global	0.8269	Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning
157	GraphSAGE GNN baseline actual #157 vs paper #280 123 places worse	0.8196 ±0.0739 best reported	actual #157 vs paper #280 123 places worse paper 0.7050 ±0.0370 best at pub: Heterophily-Aware Graph Attention Network (2023) paper 0.7050 ±0.0370 Δ at pub 0.1146
158	GAT GNN baseline actual #158 vs paper #524 366 places worse	0.8190 ±0.0540 best reported	actual #158 vs paper #524 366 places worse paper 0.4570 ±0.0630 best at pub: Restructuring Graphs for Higher Homophily via Adaptive Spectral Clustering (AAAI 2022) paper 0.4570 ±0.0630 Δ at pub 0.3620
170	GCN GNN baseline actual #170 vs paper #542 372 places worse	0.8082 ±0.0360 best reported	actual #170 vs paper #542 372 places worse paper 0.4270 ±0.0490 best at pub: Differentiable Cluster Graph Neural Network (2024) best today: Bayesian Neighborhood Adaptation for Graph Neural Networks (TMLR 2025) paper 0.4270 ±0.0490 Δ at pub 0.3745 Δ today 0.3812
201	GIN GNN baseline actual #201 vs paper #516 315 places worse	0.7800 ±0.0190 best reported	actual #201 vs paper #516 315 places worse paper 0.4680 ±0.0710 best at pub: Unifying over-smoothing and over-squashing in graph neural networks: A physics informed approach and beyond (2023) best today: Design your own universe: a physics-informed agnostic method for enhancing graph neural networks (2024) paper 0.4680 ±0.0710 Δ at pub 0.3120 Δ today 0.3120
266	GraphSAGE+het. GNN baseline	0.7220 ±0.0760	This paper
300	GraphSAGE+mix. Hyb baseline	0.6810 ±0.0650	This paper
307	GraphSAGE+hom. GNN baseline	0.6730 ±0.0410	This paper
342	Geom-GCN GNN baseline actual #342 vs paper #371 29 places worse	0.6412 best reported	actual #342 vs paper #371 29 places worse paper 0.6080 best at pub: Steering Graph Neural Networks with Pinning Control (2023) paper 0.6080 Δ at pub 0.0332
390	GAT+het. GNN baseline	0.5890 ±0.0940	This paper
484	GAT+mix. Hyb baseline	0.5050 ±0.0750	This paper
486	GIN+het. GNN baseline	0.5030 ±0.1070	This paper
491	GCN+het. GNN baseline	0.5000 ±0.0820	This paper
494	GIN+mix. Hyb baseline	0.4950 ±0.0670	This paper
499	GCN+mix. Hyb baseline	0.4890 ±0.0620	This paper
515	GAT+hom. GNN baseline	0.4700 ±0.0750	This paper
518	GIN+hom. GNN baseline	0.4680 ±0.0660	This paper
549	GCN+hom. GNN baseline	0.4140 ±0.0540	This paper

Wisconsin

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	DMD++ Hyb global	0.9190 ±0.0260	When Graph Neural Networks Meet Dynamic Mode Decomposition
2	F-GREAD-BS code ↗ GNN global	0.9059 ±0.0380	Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND code ↗
3	APPNP GNN global	0.9010 ±0.0350	When Graph Neural Networks Meet Dynamic Mode Decomposition
37	MLP code ↗ Trad global	0.8629 ±0.0450	Is Homophily a Necessity for Graph Neural Networks? code ↗
49	GraphSAGE+mix. Hyb baseline	0.8450 ±0.0110	This paper
53	GAT GNN baseline actual #53 vs paper #207 154 places worse	0.8430 ±0.0370 best reported	actual #53 vs paper #207 154 places worse paper 0.5080 ±0.0870 best at pub: Restructuring Graphs for Higher Homophily via Adaptive Spectral Clustering (AAAI 2022) paper 0.5080 ±0.0870 Δ at pub 0.3350
63	GCN GNN baseline actual #63 vs paper #199 136 places worse	0.8310 ±0.0320 best reported	actual #63 vs paper #199 136 places worse paper 0.5240 ±0.0600 best at pub: Restructuring Graphs for Higher Homophily via Adaptive Spectral Clustering (AAAI 2022) paper 0.5240 ±0.0600 Δ at pub 0.3070
69	GraphSAGE GNN baseline actual #69 vs paper #94 25 places worse	0.8118 ±0.0556 best reported	actual #69 vs paper #94 25 places worse paper 0.7710 ±0.0650 best at pub: From Node Interaction to Hop Interaction: New Effective and Scalable Graph Learning Paradigm (2022) best today: Graph Navier–Stokes Networks (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.7710 ±0.0650 Δ at pub 0.0408 Δ today 0.0408
77	GraphSAGE+hom. GNN baseline	0.8020 ±0.0520	This paper
78	GraphSAGE+het. GNN baseline	0.8000 ±0.0430	This paper
86	GIN GNN baseline actual #86 vs paper #216 130 places worse	0.7900 ±0.0060 best reported	actual #86 vs paper #216 130 places worse paper 0.4490 ±0.0680 best at pub: Unleashing the Potential of Text-attributed Graphs: Automatic Relation Decomposition via Large Language Models (2024) best today: What Can We Learn From MIMO Graph Convolutions? (IJCAI 2025) paper 0.4490 ±0.0680 Δ at pub 0.2838 Δ today 0.3410
149	SAGE GNN global	0.6480 ±0.0520	When Graph Neural Networks Meet Dynamic Mode Decomposition
187	GAT+mix. Hyb baseline	0.5530 ±0.0630	This paper
188	GAT+hom. GNN baseline	0.5490 ±0.0580	This paper
190	GCN+hom. GNN baseline	0.5450 ±0.0330	This paper
192	GCN+het. GNN baseline	0.5390 ±0.0380	This paper
195	GAT+het. GNN baseline	0.5330 ±0.0680	This paper
203	GCN+mix. Hyb baseline	0.5200 ±0.0610	This paper
204	GIN+het. GNN baseline	0.5200 ±0.0380	This paper
209	GIN+mix. Hyb baseline	0.5060 ±0.0780	This paper
214	GIN+hom. GNN baseline	0.4900 ±0.0630	This paper

Heterophily Benchmark

Roman-empire

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	Dir-Poly code ↗ GNN global	0.9451 ±0.0022	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
2	D-CDE code ↗ GNN global	0.9387 ±0.0041	Distributed-Order Fractional Graph Operating Network code ↗
3	LargeScaleNet code ↗ Hyb global	0.9358 ±0.0024	Scale-aware Message Passing For Graph Node Classification code ↗
26	GraphSAGE GNN baseline actual #26 vs paper #97 71 places worse	0.9106 ±0.0027 best reported	actual #26 vs paper #97 71 places worse paper 0.7890 ±0.0040 best at pub: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) best today: Directed Semi-Simplicial Learning with Applications to Brain Activity Decoding (2025) paper 0.7890 ±0.0040 Δ at pub 0.0684 Δ today 0.1216
64	SAGE GNN global	0.8574 ±0.0067	TANGO: Graph Neural Dynamics via Learned Energy and Tangential Flows
77	GCN GNN baseline actual #77 vs paper #193 116 places worse	0.8272 ±0.0082 best reported	actual #77 vs paper #193 116 places worse paper 0.4780 ±0.0080 best at pub: HyperAggregation: Aggregating over Graph Edges with Hypernetworks (2024) paper 0.4780 ±0.0080 Δ at pub 0.3492
83	GAT GNN baseline actual #83 vs paper #193 110 places worse	0.8145 ±0.0009 best reported	actual #83 vs paper #193 110 places worse paper 0.4710 ±0.0060 best at pub: Graph Neural Networks for Graphs with Heterophily: A Survey (2022) best today: Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax (2025) paper 0.4710 ±0.0060 Δ at pub 0.3377 Δ today 0.3435
98	GraphSAGE+mix. Hyb baseline	0.7870 ±0.0070	This paper
99	GraphSAGE+hom. GNN baseline	0.7860 ±0.0060	This paper
103	GraphSAGE+het. GNN baseline	0.7810 ±0.0070	This paper
118	GCN+mix. Hyb baseline	0.7150 ±0.0050	This paper
128	GAT+mix. Hyb baseline	0.6890 ±0.0090	This paper
135	MLP code ↗ Trad global	0.6665 ±0.0047	HyperAggregation: Aggregating over Graph Edges with Hypernetworks code ↗
140	GIN+hom. GNN baseline	0.6590 ±0.0080	This paper
151	GIN+mix. Hyb baseline	0.6400 ±0.0100	This paper
164	GCN+het. GNN baseline	0.6020 ±0.0050	This paper
170	GIN+het. GNN baseline	0.5860 ±0.0110	This paper
174	GAT+hom. GNN baseline	0.5760 ±0.0100	This paper
177	GIN GNN baseline actual #177 vs paper #192 15 places worse	0.5680 ±0.0100 best reported	actual #177 vs paper #192 15 places worse paper 0.4820 ±0.0120 best at pub: Diss-l-ECT: Dissecting Graph Data with local Euler Characteristic Transforms (ICML 2024) paper 0.4820 ±0.0120 Δ at pub 0.0860
178	GCN+hom. GNN baseline	0.5660 ±0.0060	This paper
193	GAT+het. GNN baseline	0.4800 ±0.0130	This paper

Amazon-ratings

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	UniGAP global	0.6375 ±0.0110	UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node …
2	H2GCN + HalfHop GNN global	0.5961 ±0.0100	UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node …
3	GraphSAGE + HalfHop GNN global	0.5897 ±0.0150	UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node …
12	SAGE code ↗ GNN global	0.5569 ±0.0023	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
14	GAT GNN baseline actual #14 vs paper #109 95 places worse	0.5559 ±0.0027 best reported	actual #14 vs paper #109 95 places worse paper 0.4880 ±0.0050 best at pub: Graph Neural Networks for Graphs with Heterophily: A Survey (2022) best today: Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification (2026) paper 0.4880 ±0.0050 Δ at pub 0.0029 Δ today 0.0679
15	GraphSAGE GNN baseline actual #15 vs paper #51 36 places worse	0.5540 ±0.0020 best reported	actual #15 vs paper #51 36 places worse paper 0.5270 ±0.0060 best at pub: UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node Classification Tasks (2024) paper 0.5270 ±0.0060 Δ at pub 0.0270
24	GCN GNN baseline actual #24 vs paper #105 81 places worse	0.5411 ±0.0034 best reported	actual #24 vs paper #105 81 places worse paper 0.4910 ±0.0060 best at pub: UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node Classification Tasks (2024) best today: Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification (2026) paper 0.4910 ±0.0060 Δ at pub 0.0470 Δ today 0.0501
34	GraphSAGE+mix. Hyb baseline	0.5350 ±0.0070	This paper
61	GraphSAGE+het. GNN baseline	0.5220 ±0.0060	This paper
67	GraphSAGE+hom. GNN baseline	0.5160 ±0.0050	This paper
84	MLP Trad global	0.5058	Understanding Heterophily for Graph Neural Networks
89	GAT+mix. Hyb baseline	0.5020 ±0.0050	This paper
88	GIN GNN baseline actual #88 vs paper #88	0.5020 ±0.0070	actual #88 vs paper #88 paper 0.5020 ±0.0070 prev best: Diss-l-ECT: Dissecting Graph Data with local Euler Characteristic Transforms (ICML 2024) paper 0.5020 ±0.0070 Δ at pub 0.0610
98	GCN+mix. Hyb baseline	0.4970 ±0.0080	This paper
104	GIN+hom. GNN baseline	0.4920 ±0.0070	This paper
119	GIN+het. GNN baseline	0.4820 ±0.0120	This paper
120	GIN+mix. Hyb baseline	0.4820 ±0.0150	This paper
130	GAT+hom. GNN baseline	0.4720 ±0.0070	This paper
131	GCN+hom. GNN baseline	0.4710 ±0.0050	This paper
136	GCN+het. GNN baseline	0.4640 ±0.0060	This paper
157	GAT+het. GNN baseline	0.4410 ±0.0060	This paper

Minesweeper

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	BuNN code ↗ Hyb global	0.9899 ±0.0016	Bundle Neural Networks for message diffusion on graphs code ↗
2	DC-GNN GNN global	0.9850 ±0.0021	Differentiable Cluster Graph Neural Network
3	Tango -GPS global	0.9839 ±0.0054	TANGO: Graph Neural Dynamics via Learned Energy and Tangential Flows
36	GAT GNN baseline actual #36 vs paper #154 118 places worse	0.9391 ±0.0035 best reported	actual #36 vs paper #154 118 places worse paper 0.5440 ±0.0160 best at pub: Graph Neural Networks for Graphs with Heterophily: A Survey (2022) best today: Relieving the Over-Aggregating Effect in Graph Transformers (NeurIPS 2025) paper 0.5440 ±0.0160 Δ at pub 0.3761 Δ today 0.3951
42	GraphSAGE GNN baseline actual #42 vs paper #130 88 places worse	0.9358 ±0.0005 best reported	actual #42 vs paper #130 88 places worse paper 0.7320 ±0.0170 best at pub: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) best today: Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax (2025) paper 0.7320 ±0.0170 Δ at pub 0.2031 Δ today 0.2038
46	SAGE GNN global	0.9351 ±0.0057	Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables
68	GCN GNN baseline actual #68 vs paper #155 87 places worse	0.9150 ±0.0006 best reported	actual #68 vs paper #155 87 places worse paper 0.5230 ±0.0060 best at pub: Graph Neural Networks for Graphs with Heterophily: A Survey (2022) best today: Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax (2025) paper 0.5230 ±0.0060 Δ at pub 0.3745 Δ today 0.3920
104	GIN GNN baseline actual #104 vs paper #154 50 places worse	0.8782 ±0.0034 best reported	actual #104 vs paper #154 50 places worse paper 0.5510 ±0.0320 best at pub: TopoBench: A Framework for Benchmarking Topological Deep Learning (2024) paper 0.5510 ±0.0320 Δ at pub 0.3272
127	GAT+mix. Hyb baseline	0.7510 ±0.0240	This paper
128	GCN+mix. Hyb baseline	0.7450 ±0.0230	This paper
131	GraphSAGE+mix. Hyb baseline	0.7310 ±0.0190	This paper
136	GCN+het. GNN baseline	0.7150 ±0.0180	This paper
138	GraphSAGE+het. GNN baseline	0.7100 ±0.0190	This paper
142	GraphSAGE+hom. GNN baseline	0.6840 ±0.0140	This paper
146	GIN+het. GNN baseline	0.6570 ±0.0980	This paper
151	GIN+mix. Hyb baseline	0.6050 ±0.0700	This paper
152	GIN+hom. GNN baseline	0.5820 ±0.0460	This paper
153	GCN+hom. GNN baseline	0.5730 ±0.0100	This paper
154	GAT+hom. GNN baseline	0.5550 ±0.0320	This paper
155	GAT+het. GNN baseline	0.5270 ±0.0380	This paper
156	MLP Trad global	0.5099 ±0.0147	Graph Homophily Booster: Rethinking the Role of Discrete Features on Heterophilic Graphs

Tolokers

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	Polynormer-r code ↗ GT global	0.8591 ±0.0074	Polynormer: Polynomial-Expressive Graph Transformer in Linear Time code ↗
2	DC-GNN GNN global	0.8588 ±0.0081	Differentiable Cluster Graph Neural Network
3	TANGO-GPS Hyb global	0.8566 ±0.0101	TANGO: Graph Neural Dynamics via Learned Energy and Tangential Flows
28	GAT GNN baseline actual #28 vs paper #161 133 places worse	0.8443 ±0.0100 best reported	actual #28 vs paper #161 133 places worse paper 0.5700 ±0.0270 best at pub: TopoBench: A Framework for Benchmarking Topological Deep Learning (2024) paper 0.5700 ±0.0270 Δ at pub 0.2743
50	GCN GNN baseline actual #50 vs paper #159 109 places worse	0.8364 ±0.0067 best reported	actual #50 vs paper #159 109 places worse paper 0.5800 ±0.0290 best at pub: Graph Neural Networks for Graphs with Heterophily: A Survey (2022) best today: Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables (2026) paper 0.5800 ±0.0290 Δ at pub 0.2564 Δ today 0.2564
77	GraphSAGE GNN baseline actual #77 vs paper #156 79 places worse	0.8243 ±0.0044 best reported	actual #77 vs paper #156 79 places worse paper 0.6050 ±0.0160 best at pub: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) best today: Relieving the Over-Aggregating Effect in Graph Transformers (NeurIPS 2025) paper 0.6050 ±0.0160 Δ at pub 0.2193 Δ today 0.2193
76	SAGE GNN global	0.8243 ±0.0044	Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables
89	GIN GNN baseline actual #89 vs paper #166 77 places worse	0.8072 ±0.0119 best reported	actual #89 vs paper #166 77 places worse paper 0.5330 ±0.0690 best at pub: TopoBench: A Framework for Benchmarking Topological Deep Learning (2024) paper 0.5330 ±0.0690 Δ at pub 0.2742
139	MLP code ↗ Trad global	0.7362 ±0.0020	Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax code ↗
156	GAT+mix. Hyb baseline	0.6240 ±0.0270	This paper
157	GraphSAGE+hom. GNN baseline	0.5970 ±0.0150	This paper
158	GraphSAGE+mix. Hyb baseline	0.5920 ±0.0200	This paper
159	GCN+mix. Hyb baseline	0.5870 ±0.0160	This paper
160	GCN+hom. GNN baseline	0.5770 ±0.0320	This paper
161	GraphSAGE+het. GNN baseline	0.5770 ±0.0140	This paper
162	GIN+mix. Hyb baseline	0.5630 ±0.0700	This paper
163	GAT+hom. GNN baseline	0.5570 ±0.0320	This paper
164	GCN+het. GNN baseline	0.5510 ±0.0110	This paper
165	GIN+het. GNN baseline	0.5370 ±0.0580	This paper
167	GIN+hom. GNN baseline	0.5300 ±0.0640	This paper
168	GAT+het. GNN baseline	0.5050 ±0.0050	This paper

Questions

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	NID_GAT code ↗ global	0.9703 ±0.0002	Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning code ↗
2	NID_GCN code ↗ global	0.9685 ±0.0010	Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning code ↗
3	UniGAP global	0.8421 ±0.0080	UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node …
16	GCN GNN baseline actual #16 vs paper #154 138 places worse	0.7902 ±0.0060 best reported	actual #16 vs paper #154 138 places worse paper 0.5320 ±0.0040 best at pub: UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node Classification Tasks (2024) paper 0.5320 ±0.0040 Δ at pub 0.2582
26	GAT GNN baseline actual #26 vs paper #157 131 places worse	0.7835 ±0.0116 best reported	actual #26 vs paper #157 131 places worse paper 0.5240 ±0.0040 best at pub: Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning (ICLR 2024) paper 0.5240 ±0.0040 Δ at pub 0.2595
46	GraphSAGE GNN baseline actual #46 vs paper #145 99 places worse	0.7721 ±0.0130 best reported	actual #46 vs paper #145 99 places worse paper 0.5600 ±0.0080 best at pub: UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node Classification Tasks (2024) paper 0.5600 ±0.0080 Δ at pub 0.2121
58	SAGE GNN global	0.7644 ±0.0062	Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables
96	MLP code ↗ Trad global	0.7082 ±0.0010	Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax code ↗
118	GIN GNN baseline actual #118 vs paper #159 41 places worse	0.6776 best reported	actual #118 vs paper #159 41 places worse paper 0.5180 ±0.0300 best at pub: Data Augmentation for Supervised Graph Outlier Detection via Latent Diffusion Models (LoG 2023) paper 0.5180 ±0.0300 Δ at pub 0.1596
138	GIN+mix. Hyb baseline	0.5930 ±0.0390	This paper
142	GraphSAGE+mix. Hyb baseline	0.5720 ±0.0120	This paper
145	GraphSAGE+het. GNN baseline	0.5610 ±0.0070	This paper
148	GraphSAGE+hom. GNN baseline	0.5570 ±0.0070	This paper
149	GIN+hom. GNN baseline	0.5570 ±0.0460	This paper
151	GAT+mix. Hyb baseline	0.5480 ±0.0090	This paper
155	GCN+mix. Hyb baseline	0.5310 ±0.0060	This paper
156	GAT+hom. GNN baseline	0.5300 ±0.0070	This paper
157	GCN+hom. GNN baseline	0.5280 ±0.0020	This paper
159	GCN+het. GNN baseline	0.5210 ±0.0090	This paper
160	GAT+het. GNN baseline	0.5130 ±0.0080	This paper
162	GIN+het. GNN baseline	0.5080 ±0.0260	This paper

SOTA performance (not significantly different from the best) Baseline / comparison (this paper) Global context (other papers) Best at pub — reported result worse than the best known when published (best reported in green) Best at pub — reported value improved over the best known when published Best today — OK at publication; today's leaderboard best is elsewhere