Graph Neural Aggregation-diffusion with Metastability

Kaiyuan Cui, Xinyan Wang, Zicheng Zhang, Weichen Zhao

Abstract

Continuous graph neural models based on differential equations have expanded the architecture of graph neural networks (GNNs). Due to the connection between graph diffusion and message passing, diffusion-based models have been widely studied. However, diffusion naturally drives the system towards an equilibrium state, leading to issues like over-smoothing. To this end, we propose GRADE inspired by aph ggregation- iffusion quations, which includes the delicate balance between nonlinear diffusion and aggregation induced by interaction potentials. The node representations obtained through aggregation-diffusion equations exhibit metastability, indicating that features can aggregate into multiple clusters. In addition, the dynamics within these clusters can persist for long time periods, offering the potential to alleviate over-smoothing effects. This nonlinear diffusion in our model generalizes existing diffusion-based models and establishes a connection with classical GNNs. We prove that GRADE achieves competitive performance across various benchmarks and alleviates the over-smoothing issue in GNNs evidenced by the enhanced Dirichlet energy.

Benchmark Results

Tested on 11 datasets 22 baseline values reported too low

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

Classic

Cora

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GraphSAGE+CNA code ↗ GNN global	0.9418 ±0.0033	Graph Neural Networks Need Cluster-Normalize-Activate Modules code ↗
2	GCN+CNA code ↗ GNN global	0.9366 ±0.0048	Graph Neural Networks Need Cluster-Normalize-Activate Modules code ↗
3	TransformerConv code ↗ GNN global	0.9350 ±0.0058	Graph Neural Networks Need Cluster-Normalize-Activate Modules code ↗
21	GCN GNN 0.1M baseline actual #21 vs paper #573 552 places worse	0.9004 best reported	actual #21 vs paper #573 552 places worse paper 0.8150 ±0.0130 best at pub: Understanding Heterophily for Graph Neural Networks (ICML 2024) paper 0.8150 ±0.0130 Δ at pub 0.0854
39	SAGE GNN global	0.8911 ±0.0012	Efficient Text-Attributed Graph Learning through Selective Annotation and Graph Alignment
44	GAT GNN 1.6M baseline actual #44 vs paper #550 506 places worse	0.8897 ±0.0014 best reported	actual #44 vs paper #550 506 places worse paper 0.8180 ±0.0130 best at pub: Learned Low Precision Graph Neural Networks (2020) best today: Adaptive Tokenization: On the Hop-Overpriority Problem in Tokenized Graph Learning Models (2025) paper 0.8180 ±0.0130 Δ at pub 0.0710 Δ today 0.0717
87	GRAND GNN baseline actual #87 vs paper #465 378 places worse	0.8815 best reported	actual #87 vs paper #465 378 places worse paper 0.8250 ±0.0070 best at pub: Feature Transportation Improves Graph Neural Networks (AAAI 2023) paper 0.8250 ±0.0070 Δ at pub 0.0565
224	GraphSAGE-mean GNN baseline actual #224 vs paper #688 464 places worse	0.8504 ±0.0012 best reported	actual #224 vs paper #688 464 places worse paper 0.7920 ±0.0770 best at pub: Node Representation Learning in Graph via Node-to-Neighbourhood Mutual Information Maximization (2022) Difference within reported uncertainty; not counted as underperformance. paper 0.7920 ±0.0770 Δ at pub 0.0584
240	GRAND-l GNN baseline actual #240 vs paper #340 100 places worse	0.8470 ±0.0060 best reported	actual #240 vs paper #340 100 places worse paper 0.8360 ±0.0100 best at pub: GRAND: Graph Neural Diffusion (ICML 2021) Difference within reported uncertainty; not counted as underperformance. paper 0.8360 ±0.0100 Δ at pub 0.0110
285	GRADE-GAT GNN ★ proposed	0.8420 ±0.0080	This paper
283	CGNN GNN baseline actual #283 vs paper #578 295 places worse	0.8420 ±0.0100 best reported	actual #283 vs paper #578 295 places worse paper 0.8140 ±0.0160 best at pub: Continuous Graph Neural Networks (ICML 2019) paper 0.8140 ±0.0160 Δ at pub 0.0280
305	GRADE-Gaussian Hyb ★ proposed	0.8400 ±0.0070	This paper
317	GDE GNN baseline actual #317 vs paper #704 387 places worse	0.8380 ±0.0050 best reported	actual #317 vs paper #704 387 places worse paper 0.7870 ±0.0220 best at pub: GRAND: Graph Neural Diffusion (ICML 2021) best today: Framelet Message Passing (2023) paper 0.7870 ±0.0220 Δ at pub 0.0510 Δ today 0.0510
477	GRADE-log Hyb ★ proposed	0.8240 ±0.0090	This paper
492	ACMP-GAT Hyb baseline	0.8230 ±0.0050	This paper
508	MoNet GNN baseline actual #508 vs paper #583 75 places worse	0.8220 ±0.0070 best reported	actual #508 vs paper #583 75 places worse paper 0.8130 ±0.0130 best at pub: GRAND: Graph Neural Diffusion (ICML 2021) Difference within reported uncertainty; not counted as underperformance. paper 0.8130 ±0.0130 Δ at pub 0.0090
571	GAT-ppr GNN baseline	0.8160 ±0.0030	This paper
690	GraphSage-max GNN baseline	0.7920 ±0.0770	This paper
724	MLP Trad global	0.7768	Understanding Heterophily for Graph Neural Networks

CiteSeer

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	CNA code ↗ GNN global	0.9575 ±0.0058	Graph Neural Networks Need Cluster-Normalize-Activate Modules code ↗
2	IS-GIB code ↗ Hyb global	0.9390 ±0.0187	Individual and Structural Graph Information Bottlenecks for Out-of-Distribution Generalization code ↗
3	EERM code ↗ GNN global	0.9112 ±0.0145	Individual and Structural Graph Information Bottlenecks for Out-of-Distribution Generalization code ↗
35	GAT GNN 1.6M baseline actual #35 vs paper #519 484 places worse	0.8080 ±0.0026 best reported	actual #35 vs paper #519 484 places worse paper 0.7140 ±0.0190 best at pub: Feature Expansion for Graph Neural Networks (ICML 2023) paper 0.7140 ±0.0190 Δ at pub 0.0940
44	GCN GNN 0.1M baseline actual #44 vs paper #463 419 places worse	0.8033 ±0.0088 best reported	actual #44 vs paper #463 419 places worse paper 0.7190 ±0.0190 best at pub: Gophormer: Ego-Graph Transformer for Node Classification (2021) best today: NTFormer: A Composite Node Tokenized Graph Transformer for Node Classification (2024) paper 0.7190 ±0.0190 Δ at pub 0.0753 Δ today 0.0843
61	GraphSAGE GNN global	0.7923 ±0.0053	Gophormer: Ego-Graph Transformer for Node Classification
93	GraphSAGE-mean GNN baseline actual #93 vs paper #496 403 places worse	0.7782 ±0.0015 best reported	actual #93 vs paper #496 403 places worse paper 0.7160 ±0.0200 best at pub: Node Representation Learning in Graph via Node-to-Neighbourhood Mutual Information Maximization (2022) paper 0.7160 ±0.0200 Δ at pub 0.0622
122	GRAND GNN baseline actual #122 vs paper #239 117 places worse	0.7657 best reported	actual #122 vs paper #239 117 places worse paper 0.7380 ±0.0270 best at pub: Feature Transportation Improves Graph Neural Networks (AAAI 2023) Difference within reported uncertainty; not counted as underperformance. paper 0.7380 ±0.0270 Δ at pub 0.0277
156	ACMP-GAT Hyb baseline	0.7550 ±0.0100	This paper
192	GRADE-Gaussian Hyb ★ proposed	0.7470 ±0.0100	This paper
250	GRADE-GAT GNN ★ proposed	0.7370 ±0.0130	This paper
276	GRADE-log Hyb ★ proposed	0.7350 ±0.0170	This paper
281	GRAND-l GNN baseline	0.7340 ±0.0050	This paper
350	MLP code ↗ Trad global	0.7289 ±0.0010	Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax code ↗
391	GDE GNN baseline actual #391 vs paper #474 83 places worse	0.7250 ±0.0050 best reported	actual #391 vs paper #474 83 places worse paper 0.7180 ±0.0110 best at pub: GRAND: Graph Neural Diffusion (ICML 2021) best today: Framelet Message Passing (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.7180 ±0.0110 Δ at pub 0.0070 Δ today 0.0070
498	GraphSage-max GNN baseline	0.7160 ±0.0190	This paper
540	MoNet GNN baseline	0.7120 ±0.0200	This paper
590	CGNN GNN baseline actual #590 vs paper #724 134 places worse	0.7050 ±0.0123 best reported	actual #590 vs paper #724 134 places worse paper 0.6690 ±0.0180 best at pub: GRAND: Graph Neural Diffusion (ICML 2021) best today: Gauge-Equivariant Graph Networks via Self-Interference Cancellation (2025) paper 0.6690 ±0.0180 Δ at pub 0.0120 Δ today 0.0360
671	GAT-ppr GNN baseline	0.6850 ±0.0020	This paper

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 0.1M baseline actual #18 vs paper #694 676 places worse	0.9296 ±0.0015 best reported	actual #18 vs paper #694 676 places worse paper 0.7780 ±0.0290 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 0.7780 ±0.0290 Δ at pub 0.1190 Δ today 0.1516
20	GAT GNN 1.6M baseline actual #20 vs paper #624 604 places worse	0.9233 ±0.0018 best reported	actual #20 vs paper #624 604 places worse paper 0.7870 ±0.0230 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 0.7870 ±0.0230 Δ at pub 0.1100 Δ today 0.1363
109	GRAND GNN baseline actual #109 vs paper #474 365 places worse	0.8902 best reported	actual #109 vs paper #474 365 places worse paper 0.7980 ±0.0160 best at pub: Feature Transportation Improves Graph Neural Networks (AAAI 2023) paper 0.7980 ±0.0160 Δ at pub 0.0922
116	SAGE GNN global	0.8881 ±0.0015	Efficient Text-Attributed Graph Learning through Selective Annotation and Graph Alignment
200	GraphSAGE-mean GNN baseline actual #200 vs paper #716 516 places worse	0.8715 ±0.0014 best reported	actual #200 vs paper #716 516 places worse paper 0.7740 ±0.0220 best at pub: Node Representation Learning in Graph via Node-to-Neighbourhood Mutual Information Maximization (2022) paper 0.7740 ±0.0220 Δ at pub 0.0975
210	MLP code ↗ Trad global	0.8690 ±0.0074	On the Equivalence of Graph Convolution and Mixup code ↗
412	GRAND-l GNN baseline actual #412 vs paper #614 202 places worse	0.8040 ±0.0040 best reported	actual #412 vs paper #614 202 places worse paper 0.7880 ±0.0170 best at pub: GRAND: Graph Neural Diffusion (ICML 2021) Difference within reported uncertainty; not counted as underperformance. paper 0.7880 ±0.0170 Δ at pub 0.0160
430	CGNN GNN baseline actual #430 vs paper #856 426 places worse	0.8020 ±0.0030 best reported	actual #430 vs paper #856 426 places worse paper 0.6660 ±0.0440 best at pub: GRAND: Graph Neural Diffusion (ICML 2021) best today: Framelet Message Passing (2023) paper 0.6660 ±0.0440 Δ at pub 0.1360 Δ today 0.1360
446	MoNet GNN baseline actual #446 vs paper #628 182 places worse	0.8006 ±0.0055 best reported	actual #446 vs paper #628 182 places worse paper 0.7860 ±0.0230 best at pub: Not All Neighbors Are Worth Attending to: Graph Selective Attention Networks for Semi-supervised Learning (2022) Difference within reported uncertainty; not counted as underperformance. paper 0.7860 ±0.0230 Δ at pub 0.0146
459	GDE GNN baseline actual #459 vs paper #810 351 places worse	0.7990 ±0.0030 best reported	actual #459 vs paper #810 351 places worse paper 0.7390 ±0.0370 best at pub: GRAND: Graph Neural Diffusion (ICML 2021) best today: Beltrami Flow and Neural Diffusion on Graphs (NeurIPS 2021) paper 0.7390 ±0.0370 Δ at pub 0.0600 Δ today 0.0600
509	GRADE-log Hyb ★ proposed	0.7950 ±0.0040	This paper
527	ACMP-GAT Hyb baseline	0.7940 ±0.0040	This paper
544	GRADE-GAT GNN ★ proposed	0.7930 ±0.0020	This paper
604	GRADE-Gaussian Hyb ★ proposed	0.7890 ±0.0020	This paper
717	GraphSage-max GNN baseline	0.7740 ±0.0220	This paper
751	GAT-ppr GNN baseline	0.7670 ±0.0030	This paper

Amazon-Photo

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GraphTARIF code ↗ Hyb global	0.9703 ±0.0019	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
2	GAT GNN 1.6M baseline actual #2 vs paper #503 501 places worse	0.9669 ±0.0014 best reported	actual #2 vs paper #503 501 places worse paper 0.8570 ±0.2030 best at pub: SignGT: Signed Attention-based Graph Transformer for Graph Representation Learning (2023) best today: Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.8570 ±0.2030 Δ at pub 0.0928 Δ today 0.1099
3	GAT+T code ↗ Hyb global	0.9660 ±0.0007	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
7	SAGE code ↗ GNN global	0.9643 ±0.0027	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
8	GCN GNN 0.1M baseline actual #8 vs paper #395 387 places worse	0.9632 ±0.0008 best reported	actual #8 vs paper #395 387 places worse paper 0.9120 ±0.0120 best at pub: SignGT: Signed Attention-based Graph Transformer for Graph Representation Learning (2023) best today: Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification (2026) paper 0.9120 ±0.0120 Δ at pub 0.0356 Δ today 0.0512
20	GRAND GNN baseline actual #20 vs paper #148 128 places worse	0.9350 ±0.0040	actual #20 vs paper #148 128 places worse paper 0.9350 ±0.0040 prev best: TIDE: Time Derivative Diffusion for Deep Learning on Graphs (ICML 2022) best today: Graph Wave Networks (WWW 2025) paper 0.9350 ±0.0040 Δ at pub 0.0933 Δ today 0.0229
82	CGNN GNN baseline actual #82 vs paper #378 296 places worse	0.9435 ±0.0056 best reported	actual #82 vs paper #378 296 places worse paper 0.9140 ±0.0150 best at pub: Beltrami Flow and Neural Diffusion on Graphs (NeurIPS 2021) best today: Graph Wave Networks (WWW 2025) paper 0.9140 ±0.0150 Δ today 0.0295
231	GRADE-GAT GNN ★ proposed	0.9290 ±0.0080	This paper
256	GRADE-Gaussian Hyb ★ proposed	0.9270 ±0.0070	This paper
296	GDE GNN baseline	0.9240 ±0.0200	This paper
310	GRAND-l GNN baseline	0.9230 ±0.0090	This paper
378	GraphSAGE-mean GNN baseline actual #378 vs paper #378	0.9143 ±0.0130 best reported	actual #378 vs paper #378 paper 0.9140 ±0.0130 best at pub: Building Shortcuts between Distant Nodes with Biaffine Mapping for Graph Convolutional Networks (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.9140 ±0.0130 Δ at pub 0.0003
381	GraphSage-max GNN baseline actual #381 vs paper #381	0.9140 ±0.0130	actual #381 vs paper #381 paper 0.9140 ±0.0130 prev best: ACMP: Allen-Cahn Message Passing with Attractive and Repulsive Forces for Graph Neural Networks (ICLR 2022) Improvement within reported uncertainty; not highlighted. paper 0.9140 ±0.0130 Δ at pub 0.0100
390	MLP code ↗ Trad global	0.9128 ±0.0063	Clarify Confused Nodes via Separated Learning code ↗
393	MoNet GNN baseline actual #393 vs paper #395 2 places worse	0.9124 ±0.0130 best reported	actual #393 vs paper #395 2 places worse paper 0.9120 ±0.0230 best at pub: Building Shortcuts between Distant Nodes with Biaffine Mapping for Graph Convolutional Networks (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.9120 ±0.0230 Δ at pub 0.0004
400	ACMP-GAT Hyb baseline	0.9110 ±0.0070	This paper
403	GAT-ppr GNN baseline	0.9090 ±0.0030	This paper
439	GRADE-log Hyb ★ proposed	0.9030 ±0.0100	This paper

Amazon-Computers

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GraphTARIF code ↗ Hyb global	0.9461 ±0.0017	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
2	GCN GNN 0.1M baseline actual #2 vs paper #389 387 places worse	0.9412 ±0.0008 best reported	actual #2 vs paper #389 387 places worse paper 0.8260 ±0.0240 best at pub: Clarify Confused Nodes via Separated Learning (2023) best today: Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification (2026) paper 0.8260 ±0.0240 Δ at pub 0.0712 Δ today 0.1152
3	GAT GNN 1.6M baseline actual #3 vs paper #442 439 places worse	0.9398 ±0.0022 best reported	actual #3 vs paper #442 439 places worse paper 0.7800 ±0.1900 best at pub: SignGT: Signed Attention-based Graph Transformer for Graph Representation Learning (2023) best today: Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.7800 ±0.1900 Δ at pub 0.1278 Δ today 0.1598
8	SAGE code ↗ GNN global	0.9351 ±0.0006	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
144	GRAND GNN baseline actual #144 vs paper #346 202 places worse	0.8410 ±0.0050	actual #144 vs paper #346 202 places worse paper 0.8410 ±0.0050 prev best: TIDE: Time Derivative Diffusion for Deep Learning on Graphs (ICML 2022) best today: Graph Wave Networks (WWW 2025) paper 0.8410 ±0.0050 Δ at pub 0.1114 Δ today 0.0527
195	CGNN GNN baseline actual #195 vs paper #428 233 places worse	0.8831 ±0.0065 best reported	actual #195 vs paper #428 233 places worse paper 0.8030 ±0.0200 best at pub: Beltrami Flow and Neural Diffusion on Graphs (NeurIPS 2021) best today: Graph Wave Networks (WWW 2025) paper 0.8030 ±0.0200 Δ today 0.0801
296	GRADE-log Hyb ★ proposed	0.8570 ±0.0150	This paper
301	GRADE-GAT GNN ★ proposed	0.8540 ±0.0110	This paper
300	GAT-ppr GNN baseline	0.8540 ±0.0010	This paper
320	MLP code ↗ Trad global	0.8463	GraphFM: A generalist graph transformer that learns transferable representations across diverse … code ↗
321	GRADE-Gaussian Hyb ★ proposed	0.8460 ±0.0140	This paper
331	ACMP-GAT Hyb baseline	0.8440 ±0.0160	This paper
360	GRAND-l GNN baseline	0.8370 ±0.0120	This paper
367	MoNet GNN baseline	0.8350 ±0.0220	This paper
385	GDE GNN baseline actual #385 vs paper #407 22 places worse	0.8290 ±0.0060 best reported	actual #385 vs paper #407 22 places worse paper 0.8190 ±0.0060 best at pub: Beltrami Flow and Neural Diffusion on Graphs (NeurIPS 2021) best today: Graph Navier–Stokes Networks (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.8190 ±0.0060 Δ at pub 0.0100 Δ today 0.0100
392	GraphSAGE-mean GNN baseline	0.8240 ±0.0180	This paper
393	GraphSage-max GNN baseline	0.8240 ±0.0180	This paper

Coauthor-CS

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	DoG code ↗ GNN global	0.9690 ±0.0030	Diffusion on Graph: Augmentation of Graph Structure for Node Classification code ↗
2	GWN-fa code ↗ Hyb global	0.9667 ±0.0026	Graph Wave Networks code ↗
3	GWN-sym code ↗ GNN global	0.9666 ±0.0026	Graph Wave Networks code ↗
4	CGNN GNN baseline actual #4 vs paper #107 103 places worse	0.9621 ±0.0032 best reported	actual #4 vs paper #107 103 places worse paper 0.9230 ±0.0020 best at pub: Beltrami Flow and Neural Diffusion on Graphs (NeurIPS 2021) best today: Graph Wave Networks (WWW 2025) paper 0.9230 ±0.0020 Δ today 0.0391
6	GraphSAGE code ↗ GNN global	0.9593 ±0.0025	Graph Wave Networks code ↗
10	GRAND GNN baseline actual #10 vs paper #161 151 places worse	0.9577 ±0.0028 best reported	actual #10 vs paper #161 151 places worse paper 0.9020 ±0.0030 best at pub: TIDE: Time Derivative Diffusion for Deep Learning on Graphs (ICML 2022) best today: Graph Wave Networks (WWW 2025) paper 0.9020 ±0.0030 Δ at pub 0.0039 Δ today 0.0557
12	GCN GNN 0.1M baseline actual #12 vs paper #148 136 places worse	0.9514 ±0.0025 best reported	actual #12 vs paper #148 136 places worse paper 0.9110 ±0.0050 best at pub: Unifying Graph Convolutional Neural Networks and Label Propagation (2020) best today: Graph Wave Networks (WWW 2025) paper 0.9110 ±0.0050 Δ at pub 0.0250 Δ today 0.0404
21	GAT GNN 1.6M baseline actual #21 vs paper #156 135 places worse	0.9437 ±0.0028 best reported	actual #21 vs paper #156 135 places worse paper 0.9050 ±0.0060 best at pub: Unifying Graph Convolutional Neural Networks and Label Propagation (2020) best today: Graph Wave Networks (WWW 2025) paper 0.9050 ±0.0060 Δ at pub 0.0330 Δ today 0.0387
85	GRAND-l GNN baseline	0.9290 ±0.0040	This paper
88	MLP Trad global	0.9280 ±0.0020	Revisiting the Necessity of Graph Learning and Common Graph Benchmarks
97	GRADE-Gaussian Hyb ★ proposed	0.9270 ±0.0020	This paper
120	GRADE-GAT GNN ★ proposed	0.9210 ±0.0010	This paper
132	ACMP-GAT Hyb baseline	0.9180 ±0.0010	This paper
135	GDE GNN baseline	0.9160 ±0.0010	This paper
139	GraphSAGE-mean GNN baseline actual #139 vs paper #139	0.9132 ±0.0290 best reported	actual #139 vs paper #139 paper 0.9130 ±0.0280 best at pub: Building Shortcuts between Distant Nodes with Biaffine Mapping for Graph Convolutional Networks (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.9130 ±0.0280 Δ at pub 0.0002
143	GRADE-log Hyb ★ proposed	0.9130 ±0.0040	This paper
140	GAT-ppr GNN baseline	0.9130 ±0.0010	This paper
142	GraphSage-max GNN baseline actual #142 vs paper #142	0.9130 ±0.0280	actual #142 vs paper #142 paper 0.9130 ±0.0280 prev best: ACMP: Allen-Cahn Message Passing with Attractive and Repulsive Forces for Graph Neural Networks (ICLR 2022) paper 0.9130 ±0.0280 Δ at pub 0.0630
155	MoNet GNN baseline actual #155 vs paper #155	0.9082 ±0.0160 best reported	actual #155 vs paper #155 paper 0.9080 ±0.0060 best at pub: Building Shortcuts between Distant Nodes with Biaffine Mapping for Graph Convolutional Networks (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.9080 ±0.0060 Δ at pub 0.0002

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 ↗
5	ACM-GCN GNN baseline actual #5 vs paper #93 88 places worse	0.9557 ±0.0360 best reported	actual #5 vs paper #93 88 places worse paper 0.8780 ±0.0440 best at pub: From Node Interaction to Hop Interaction: New Effective and Scalable Graph Learning Paradigm (2022) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) paper 0.8780 ±0.0440 Δ at pub 0.0004 Δ today 0.0777
34	MLP Trad global	0.9230 ±0.0070	Design your own universe: a physics-informed agnostic method for enhancing graph neural networks
39	GGCN GNN baseline actual #39 vs paper #153 114 places worse	0.9213 ±0.0305 best reported	actual #39 vs paper #153 114 places worse paper 0.8480 ±0.0450 best at pub: Understanding convolution on graphs via energies (TMLR 2022) best today: Graph Wave Networks (WWW 2025) paper 0.8480 ±0.0450 Δ at pub 0.0006 Δ today 0.0733
40	H2GCN GNN baseline actual #40 vs paper #153 113 places worse	0.9189 ±0.0393 best reported	actual #40 vs paper #153 113 places worse paper 0.8480 ±0.0720 best at pub: How Curvature Enhance the Adaptation Power of Framelet GCNs (2023) best today: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.8480 ±0.0720 Δ at pub 0.0110 Δ today 0.0709
53	GraphSAGE code ↗ GNN global	0.9000 ±0.0365	Graph Neural Networks Need Cluster-Normalize-Activate Modules code ↗
75	GREAD-BS Hyb baseline actual #75 vs paper #76 1 place worse	0.8892 ±0.0372 best reported	actual #75 vs paper #76 1 place worse paper 0.8890 ±0.0370 best today: Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND (ICLR 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.8890 ±0.0370 Δ today 0.0002
85	GRADE-GAT GNN ★ proposed	0.8830 ±0.0350	This paper
115	ACMP-GCN Hyb baseline	0.8620 ±0.0300	This paper
141	GloGNN GNN baseline actual #141 vs paper #164 23 places worse	0.8524 ±0.0490 best reported	actual #141 vs paper #164 23 places worse paper 0.8430 ±0.0410 best at pub: Finding Global Homophily in Graph Neural Networks When Meeting Heterophily (ICML 2022) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.8430 ±0.0410 Δ at pub 0.0002 Δ today 0.0094
306	GRAND-l GNN baseline	0.7570 ±0.0730	This paper
325	LINKX GNN baseline	0.7460 ±0.0830	This paper
339	GDE GNN baseline	0.7410 ±0.0690	This paper
376	CGNN GNN baseline actual #376 vs paper #376	0.7140 ±0.0410	actual #376 vs paper #376 paper 0.7140 ±0.0410 prev best: Understanding convolution on graphs via energies (TMLR 2022) Improvement within reported uncertainty; not highlighted. paper 0.7140 ±0.0410 Δ at pub 0.0005
415	Geom-GCN GNN baseline actual #415 vs paper #425 10 places worse	0.6760 best reported	actual #415 vs paper #425 10 places worse paper 0.6670 ±0.0270 best at pub: Non-Local Graph Neural Networks (2020) Difference within reported uncertainty; not counted as underperformance. paper 0.6670 ±0.0270 Δ at pub 0.0090

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 #91 89 places worse	0.9480 ±0.0380 best reported	actual #2 vs paper #91 89 places worse paper 0.8510 ±0.0610 best at pub: From Node Interaction to Hop Interaction: New Effective and Scalable Graph Learning Paradigm (2022) best today: Heterophily-informed Message Passing (TMLR 2025) paper 0.8510 ±0.0610 Δ at pub 0.0004 Δ today 0.0970
3	RF-GCN GNN global	0.9429	Simplifying Graph Convolutional Networks with Redundancy-Free Neighbors
15	GGCN GNN baseline actual #15 vs paper #82 67 places worse	0.9230 ±0.0367 best reported	actual #15 vs paper #82 67 places worse paper 0.8560 ±0.0660 best at pub: Understanding convolution on graphs via energies (TMLR 2022) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.8560 ±0.0660 Δ at pub 0.0008 Δ today 0.0670
24	MLP Trad global	0.9136 ±0.0070	Node-oriented Spectral Filtering for Graph Neural Networks
60	GREAD-BS Hyb baseline	0.8650 ±0.0720	This paper
68	H2GCN GNN baseline actual #68 vs paper #136 68 places worse	0.8623 ±0.0471 best reported	actual #68 vs paper #136 68 places worse paper 0.8270 ±0.0520 best at pub: Design your own universe: a physics-informed agnostic method for enhancing graph neural networks (2024) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.8270 ±0.0520 Δ at pub 0.0240 Δ today 0.0353
74	GloGNN GNN baseline actual #74 vs paper #122 48 places worse	0.8595 ±0.0510 best reported	actual #74 vs paper #122 48 places worse paper 0.8350 ±0.0420 best at pub: GPNet: Simplifying Graph Neural Networks via Multi-channel Geometric Polynomials (2022) best today: Aggregation-aware MLP: An Unsupervised Approach for Graph Message-passing (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.8350 ±0.0420 Δ at pub 0.0245 Δ today 0.0245
83	ACMP-GCN Hyb baseline	0.8540 ±0.0700	This paper
123	LINKX GNN baseline actual #123 vs paper #204 81 places worse	0.8346 ±0.0061 best reported	actual #123 vs paper #204 81 places worse paper 0.7780 ±0.0580 best at pub: GPNet: Simplifying Graph Neural Networks via Multi-channel Geometric Polynomials (2022) best today: Differentiable Cluster Graph Neural Network (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.7780 ±0.0580 Δ at pub 0.0004 Δ today 0.0566
127	GRADE-GAT GNN ★ proposed	0.8330 ±0.0700	This paper
138	SAGE GNN global	0.8269	Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning
143	GDE GNN baseline actual #143 vs paper #145 2 places worse	0.8243 ±0.0707 best reported	actual #143 vs paper #145 2 places worse paper 0.8240 ±0.0710 best today: Graph Navier–Stokes Networks (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.8240 ±0.0710 Δ today 0.0003
148	GRAND-l GNN baseline	0.8220 ±0.0710	This paper
319	CGNN GNN baseline actual #319 vs paper #319	0.6622 ±0.0769 best reported	actual #319 vs paper #319 paper 0.6620 ±0.0770 best at pub: Understanding convolution on graphs via energies (TMLR 2022) best today: Graph Navier–Stokes Networks (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.6620 ±0.0770 Δ at pub 0.0002 Δ today 0.0002
342	Geom-GCN GNN baseline actual #342 vs paper #375 33 places worse	0.6412 best reported	actual #342 vs paper #375 33 places worse paper 0.6050 ±0.0360 best at pub: Steering Graph Neural Networks with Pinning Control (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.6050 ±0.0360 Δ at pub 0.0362

Actor

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	DiGGR Hyb global	0.4535 ±0.0353	Disentangled Generative Graph Representation Learning
2	TRIGON code ↗ Hyb global	0.4402 ±0.0028	Dynamic Triangulation-Based Graph Rewiring for Graph Neural Networks code ↗
3	DoG code ↗ ext data GNN global	0.4320 ±0.0120	Diffusion on Graph: Augmentation of Graph Structure for Node Classification code ↗
16	ACM-GCN GNN baseline actual #16 vs paper #171 155 places worse	0.4198 ±0.0089 best reported	actual #16 vs paper #171 155 places worse paper 0.3630 ±0.0110 best at pub: ES-GNN: Generalizing Graph Neural Networks Beyond Homophily with Edge Splitting (2022) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) paper 0.3630 ±0.0110 Δ at pub 0.0030 Δ today 0.0568
42	MLP Trad global	0.4026 ±0.0108	Heterophily-Aware Graph Attention Network
58	GloGNN GNN baseline actual #58 vs paper #117 59 places worse	0.3959 ±0.0029 best reported	actual #58 vs paper #117 59 places worse paper 0.3730 ±0.0130 best at pub: SignGT: Signed Attention-based Graph Transformer for Graph Representation Learning (2023) paper 0.3730 ±0.0130 Δ at pub 0.0229
62	GGCN GNN baseline actual #62 vs paper #113 51 places worse	0.3929 ±0.0048 best reported	actual #62 vs paper #113 51 places worse paper 0.3750 ±0.0150 best at pub: Feature Transportation Improves Graph Neural Networks (AAAI 2023) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.3750 ±0.0150 Δ at pub 0.0031 Δ today 0.0179
68	GraphSAGE code ↗ GNN global	0.3899 ±0.0085	A Quasi-Wasserstein Loss for Learning Graph Neural Networks code ↗
71	H2GCN GNN baseline actual #71 vs paper #198 127 places worse	0.3885 ±0.0117 best reported	actual #71 vs paper #198 127 places worse paper 0.3570 ±0.0100 best at pub: ES-GNN: Generalizing Graph Neural Networks Beyond Homophily with Edge Splitting (2022) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) paper 0.3570 ±0.0100 Δ at pub 0.0270 Δ today 0.0315
87	Geom-GCN GNN baseline actual #87 vs paper #310 223 places worse	0.3810 best reported	actual #87 vs paper #310 223 places worse paper 0.3150 ±0.0110 best at pub: Beyond Low-frequency Information in Graph Convolutional Networks (AAAI 2021) paper 0.3150 ±0.0110 Δ at pub 0.0660
94	GREAD-BS Hyb baseline	0.3790 ±0.0120	This paper
143	GRADE-GAT GNN ★ proposed	0.3680 ±0.0110	This paper
180	LINKX GNN baseline	0.3610 ±0.0150	This paper
189	CGNN GNN baseline actual #189 vs paper #190 1 place worse	0.3595 ±0.0086 best reported	actual #189 vs paper #190 1 place worse paper 0.3590 ±0.0090 best at pub: Understanding convolution on graphs via energies (TMLR 2022) best today: Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND (ICLR 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.3590 ±0.0090 Δ at pub 0.0005 Δ today 0.0005
209	ACMP-GCN Hyb baseline	0.3560 ±0.0110	This paper
219	GDE GNN baseline	0.3540 ±0.0130	This paper
231	GRAND-l GNN baseline	0.3510 ±0.0120	This paper

Chameleon

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	Hetero-S (GAT 8-layer) code ↗ Hyb global	0.8693	The Heterophilic Snowflake Hypothesis: Training and Empowering GNNs for Heterophilic Graphs code ↗
2	Dir-GNN+CNA code ↗ GNN global	0.8586 ±0.0180	Graph Neural Networks Need Cluster-Normalize-Activate Modules code ↗
3	GAT code ↗ GNN global	0.8560	The Heterophilic Snowflake Hypothesis: Training and Empowering GNNs for Heterophilic Graphs code ↗
28	ACM-GCN GNN baseline actual #28 vs paper #76 48 places worse	0.7476 ±0.0220 best reported	actual #28 vs paper #76 48 places worse paper 0.6690 ±0.0190 best at pub: From Node Interaction to Hop Interaction: New Effective and Scalable Graph Learning Paradigm (2022) best today: Directed Homophily-Aware Graph Neural Network (2025) paper 0.6690 ±0.0190 Δ at pub 0.0003 Δ today 0.0786
42	GREAD-BS Hyb baseline actual #42 vs paper #42	0.7142 ±0.0178 best reported	actual #42 vs paper #42 paper 0.7140 ±0.0130 best today: Latent-Hysteresis Graph ODEs: Modeling Coupled Topology-Feature Evolution via Continuous Phase Transitions (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.7140 ±0.0130 Δ today 0.0002
45	GloGNN GNN baseline actual #45 vs paper #56 11 places worse	0.7133 ±0.0242 best reported	actual #45 vs paper #56 11 places worse paper 0.6980 ±0.0240 best at pub: GPNet: Simplifying Graph Neural Networks via Multi-channel Geometric Polynomials (2022) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.6980 ±0.0240 Δ at pub 0.0141 Δ today 0.0153
48	GGCN GNN baseline actual #48 vs paper #48	0.7114 ±0.0184 best reported	actual #48 vs paper #48 paper 0.7110 ±0.0180 best at pub: GPNet: Simplifying Graph Neural Networks via Multi-channel Geometric Polynomials (2022) best today: Scalable Graph Compressed Convolutions (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.7110 ±0.0180 Δ at pub 0.0004 Δ today 0.0004
52	GCN GNN global	0.7033	Understanding Heterophily for Graph Neural Networks
64	LINKX GNN baseline actual #64 vs paper #64	0.6842 ±0.0132 best reported	actual #64 vs paper #64 paper 0.6840 ±0.0130 best at pub: GPNet: Simplifying Graph Neural Networks via Multi-channel Geometric Polynomials (2022) best today: Directed Homophily-Aware Graph Neural Network (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.6840 ±0.0130 Δ at pub 0.0002 Δ today 0.0002
124	Geom-GCN GNN baseline actual #124 vs paper #137 13 places worse	0.6090 best reported	actual #124 vs paper #137 13 places worse paper 0.6000 ±0.0280 best at pub: Non-Local Graph Neural Networks (2020) best today: Beyond Smoothing: Unsupervised Graph Representation Learning with Edge Heterophily Discriminating (AAAI 2022) Difference within reported uncertainty; not counted as underperformance. paper 0.6000 ±0.0280 Δ at pub 0.0090 Δ today 0.0090
133	H2GCN GNN baseline actual #133 vs paper #137 4 places worse	0.6030 ±0.0231 best reported	actual #133 vs paper #137 4 places worse paper 0.6010 ±0.0210 best at pub: Optimization-Induced Graph Implicit Nonlinear Diffusion (ICML 2022) Difference within reported uncertainty; not counted as underperformance. paper 0.6010 ±0.0210 Δ at pub 0.0020
143	CGNN GNN baseline actual #143 vs paper #215 72 places worse	0.5910 ±0.0078 best reported	actual #143 vs paper #215 72 places worse paper 0.4690 ±0.0170 best today: Gauge-Equivariant Graph Networks via Self-Interference Cancellation (2025) paper 0.4690 ±0.0170 Δ today 0.1220
165	GRADE-GAT GNN ★ proposed	0.5510 ±0.0240	This paper
167	ACMP-GCN Hyb baseline	0.5470 ±0.0250	This paper
169	MLP Trad global	0.5429	Understanding Heterophily for Graph Neural Networks
181	GRAND-l GNN baseline	0.5210 ±0.0300	This paper
208	GDE GNN baseline	0.4780 ±0.0210	This paper

Squirrel

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	IBG-NN GNN global	0.7763 ±0.0179	Efficient Learning on Large Graphs using a Densifying Regularity Lemma
2	Dir-GNN+CNA code ↗ GNN global	0.7747 ±0.0128	Graph Neural Networks Need Cluster-Normalize-Activate Modules code ↗
3	DHGNN GNN global	0.7684 ±0.0161	Directed Homophily-Aware Graph Neural Network
19	GloGNN GNN baseline actual #19 vs paper #58 39 places worse	0.7121 ±0.0184 best reported	actual #19 vs paper #58 39 places worse paper 0.5750 ±0.0130 best at pub: GPNet: Simplifying Graph Neural Networks via Multi-channel Geometric Polynomials (2022) best today: Directed Homophily-Aware Graph Neural Network (2025) paper 0.5750 ±0.0130 Δ at pub 0.0038 Δ today 0.1371
25	ACM-GCN GNN baseline actual #25 vs paper #75 50 places worse	0.6740 ±0.0221 best reported	actual #25 vs paper #75 50 places worse paper 0.5440 ±0.0190 best at pub: From Node Interaction to Hop Interaction: New Effective and Scalable Graph Learning Paradigm (2022) best today: Directed Homophily-Aware Graph Neural Network (2025) paper 0.5440 ±0.0190 Δ today 0.1300
39	H2GCN GNN baseline actual #39 vs paper #203 164 places worse	0.6190 ±0.0140 best reported	actual #39 vs paper #203 164 places worse paper 0.3640 ±0.0180 best at pub: Optimization-Induced Graph Implicit Nonlinear Diffusion (ICML 2022) best today: Efficient Learning on Large Graphs using a Densifying Regularity Lemma (2025) paper 0.3640 ±0.0180 Δ at pub 0.0435 Δ today 0.2550
40	LINKX GNN baseline actual #40 vs paper #40	0.6181 ±0.0180 best reported	actual #40 vs paper #40 paper 0.6180 ±0.0180 best at pub: GPNet: Simplifying Graph Neural Networks via Multi-channel Geometric Polynomials (2022) best today: Directed Homophily-Aware Graph Neural Network (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.6180 ±0.0180 Δ at pub 0.0001 Δ today 0.0001
46	GCN GNN global	0.6067	Understanding Heterophily for Graph Neural Networks
52	GREAD-BS Hyb baseline actual #52 vs paper #52	0.5922 ±0.0144 best reported	actual #52 vs paper #52 paper 0.5920 ±0.0140 best today: Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND (ICLR 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.5920 ±0.0140 Δ today 0.0002
69	GGCN GNN baseline actual #69 vs paper #71 2 places worse	0.5567 ±0.0187 best reported	actual #69 vs paper #71 2 places worse paper 0.5510 ±0.0150 best at pub: GPNet: Simplifying Graph Neural Networks via Multi-channel Geometric Polynomials (2022) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.5510 ±0.0150 Δ at pub 0.0007 Δ today 0.0057
115	GRADE-GAT GNN ★ proposed	0.4540 ±0.0190	This paper
163	ACMP-GCN Hyb baseline	0.4010 ±0.0150	This paper
172	MLP code ↗ Trad global	0.3930 ±0.0079	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
179	Geom-GCN GNN baseline actual #179 vs paper #182 3 places worse	0.3832 ±0.0092 best reported	actual #179 vs paper #182 3 places worse paper 0.3810 ±0.0090 best at pub: Feature Transportation Improves Graph Neural Networks (AAAI 2023) best today: Graph Neural Reaction Diffusion Models Submitted to the editors June 2023. Accepted in March 2024. (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.3810 ±0.0090 Δ at pub 0.0022 Δ today 0.0022
180	GRAND-l GNN baseline	0.3830 ±0.0170	This paper
207	GDE GNN baseline	0.3590 ±0.0190	This paper
228	CGNN GNN baseline actual #228 vs paper #242 14 places worse	0.3440 ±0.0097 best reported	actual #228 vs paper #242 14 places worse paper 0.2920 ±0.0110 best today: Gauge-Equivariant Graph Networks via Self-Interference Cancellation (2025) paper 0.2920 ±0.0110 Δ today 0.0520

SOTA performance (not significantly different from the best) Proposed method (this paper) 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 Improved at pub — later paper surpassed this result Best today — OK at publication; today's leaderboard best is elsewhere