GNNBenchmark

Vijay Prakash Dwivedi, Chaitanya K. Joshi, T. Laurent, Yoshua Bengio, X. Bresson

Abstract

In the last few years, graph neural networks (GNNs) have become the standard toolkit for analyzing and learning from data on graphs. This emerging field has witnessed an extensive growth of promising techniques that have been applied with success to computer science, mathematics, biology, physics and chemistry. But for any successful field to become mainstream and reliable, benchmarks must be developed to quantify progress. This led us in March 2020 to release a benchmark framework that i) comprises of a diverse collection of mathematical and real-world graphs, ii) enables fair model comparison with the same parameter budget to identify key architectures, iii) has an open-source, easy-to-use and reproducible code infrastructure, and iv) is flexible for researchers to experiment with new theoretical ideas. As of December 2022, the GitHub repository. has reached 2,000 stars and 380 forks, which demonstrates the utility of the proposed open-source framework through the wide usage by the GNN community. In this paper, we present an updated version of our benchmark with a concise presentation of the aforementioned framework characteristics, an additional medium-sized molecular dataset AQSOL, similar to the popular ZINC, but with a real-world measured chemical target, and discuss how this framework can be leveraged to explore new GNN designs and insights. As a proof of value of our benchmark, we study the case of graph positional encoding (PE) in GNNs, which was introduced with this benchmark and has since spurred interest of exploring more powerful PE for Transformers and GNNs in a robust experimental setting.

Benchmark Results

Tested on 11 datasets 18 baseline values reported too low

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

Classic

WikiCS

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GCN GNN baseline actual #1 vs paper #193 192 places worse	0.9302 ±0.0011 best reported	actual #1 vs paper #193 192 places worse paper 0.7747 ±0.0085 best at pub: Augmentation-Free Graph Contrastive Learning with Performance Guarantee (2022) paper 0.7747 ±0.0085 Δ at pub 0.1555
2	RSE + RGCN (70b) ext data global	0.8852 ±0.0019	Unleashing the Potential of Text-attributed Graphs: Automatic Relation Decomposition via Large …
3	LEMP+T ext data Hyb global	0.8825	Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning
9	SAGE GNN global	0.8771	Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning
31	GAT GNN baseline actual #31 vs paper #209 178 places worse	0.8533 best reported	actual #31 vs paper #209 178 places worse paper 0.7691 ±0.0082 best at pub: Learning Graph Augmentations to Learn Graph Representations (2022) best today: Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning (2025) paper 0.7691 ±0.0082 Δ at pub 0.0079 Δ today 0.0842
54	GIN GNN baseline actual #54 vs paper #218 164 places worse	0.8303 ±0.0021 best reported	actual #54 vs paper #218 164 places worse paper 0.7586 ±0.0058 best at pub: Learning Graph Augmentations to Learn Graph Representations (2022) best today: Unleashing the Potential of Text-attributed Graphs: Automatic Relation Decomposition via Large Language Models (2024) paper 0.7586 ±0.0058 Δ at pub 0.0004 Δ today 0.0717
69	GraphSAGE GNN baseline actual #69 vs paper #236 167 places worse	0.8146 ±0.0010 best reported	actual #69 vs paper #236 167 places worse paper 0.7477 ±0.0095 best today: Energy-Balanced Hyperspherical Graph Representation Learning via Structural Binding and Entropic Dispersion (2025) paper 0.7477 ±0.0095 Δ today 0.0669
193	MoNet-PE GNN 0.1M baseline	0.7748 ±0.0071	This paper
194	MoNet GNN baseline	0.7743 ±0.0067	This paper
206	vanilla GCN GNN baseline	0.7710 ±0.0083	This paper
237	MLP Trad baseline actual #237 vs paper #306 69 places worse	0.7469 ±0.0085 best reported	actual #237 vs paper #306 69 places worse paper 0.5945 ±0.0233 best at pub: Augmentation-Free Graph Contrastive Learning with Performance Guarantee (2022) best today: View Space: Learning Representation across Arbitrary Graphs (ICML 2025) paper 0.5945 ±0.0233 Δ at pub 0.1253 Δ today 0.1524

AQSOL

Global leaderboard →

#	Model	MAE ↓	Paper
1	Ours(100K) Hyb 0.1M global	0.9200 ±0.0090	Rewiring with Positional Encodings for Graph Neural Networks
3	GatedGCN-E-PE ext data GNN baseline	0.9960 ±0.0080	This paper
2	GatedGCN-PE/E(500K) GNN 0.5M global	0.9960 ±0.0080	Rewiring with Positional Encodings for Graph Neural Networks
4	3WLGNN-E GNN baseline	1.0420 ±0.0640	This paper
8	3WLGNN GNN baseline	1.1540 ±0.0500	This paper
10	GAT GNN baseline actual #10 vs paper #19 9 places worse	1.2400 ±0.0110 best reported	actual #10 vs paper #19 9 places worse paper 1.4410 ±0.0230 best today: Graph Alignment for Benchmarking Graph Neural Networks and Learning Positional Encodings (ICML 2025) paper 1.4410 ±0.0230 Δ today 0.2010
13	GCN GNN baseline actual #13 vs paper #17 4 places worse	1.2800 ±0.0110 best reported	actual #13 vs paper #17 4 places worse paper 1.3720 ±0.0200 best today: Graph Alignment for Benchmarking Graph Neural Networks and Learning Positional Encodings (ICML 2025) paper 1.3720 ±0.0200 Δ today 0.0920
14	GatedGCN-E GNN baseline	1.2950 ±0.0160	This paper
17	GatedGCN GNN baseline	1.3520 ±0.0340	This paper
18	MoNet GNN baseline	1.3950 ±0.0270	This paper
19	GraphSAGE GNN baseline	1.4310 ±0.0100	This paper
20	vanilla GCN GNN baseline	1.4830 ±0.0140	This paper
22	MLP Trad baseline	1.7440 ±0.0160	This paper
23	GIN GNN baseline	1.8940 ±0.0240	This paper
24	RingGNN-E GNN baseline	3.7690 ±1.0120	This paper
25	RingGNN GNN baseline	20.2640 ±7.5490	This paper

TU Dortmund

PROTEINS

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	MSH-GNN GNN global	0.9410 ±0.0330	Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing
2	GRDL GNN global	0.8260 ±0.0120	Graph Classification via Reference Distribution Learning: Theory and Practice
3	GMN code ↗ Hyb global	0.8225	Memory-Based Graph Networks code ↗
32	GIN GNN baseline actual #32 vs paper #308 276 places worse	0.7880 ±0.0410 best reported	actual #32 vs paper #308 276 places worse paper 0.7412 ±0.0336 best at pub: Breaking the Limits of Message Passing Graph Neural Networks (ICML 2021) best today: Permutation-Invariant graph partitioning: How graph neural networks capture structural interactions? (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.7412 ±0.0336 Δ at pub 0.0198 Δ today 0.0468
49	GAT GNN baseline actual #49 vs paper #131 82 places worse	0.7628 ±0.0241	actual #49 vs paper #131 82 places worse paper 0.7628 ±0.0241 prev best: Breaking the Limits of Message Passing Graph Neural Networks (ICML 2021) Improvement within reported uncertainty; not highlighted. paper 0.7628 ±0.0241 Δ at pub 0.0038
67	SAGE GNN global	0.7726 ±0.0228	Semi-Supervised Hierarchical Graph Classification
95	GatedGCN GNN baseline	0.7672 ±0.0311	This paper
111	MoNet GNN baseline	0.7645 ±0.0289	This paper
144	vanilla GCN GNN baseline	0.7610 ±0.0241	This paper
149	MLP Trad baseline actual #149 vs paper #163 14 places worse	0.7582 ±0.0291	actual #149 vs paper #163 14 places worse paper 0.7582 ±0.0291 prev best: Breaking the Limits of Message Passing Graph Neural Networks (ICML 2021) Improvement within reported uncertainty; not highlighted. paper 0.7582 ±0.0291 Δ at pub 0.0152
155	GraphSAGE GNN baseline actual #155 vs paper #190 35 places worse	0.7556 ±0.0191	actual #155 vs paper #190 35 places worse paper 0.7556 ±0.0191 prev best: Graph Convolutional Networks with EigenPooling (KDD 2019) paper 0.7556 ±0.0191 Δ at pub 0.0096
472	RingGNN GNN baseline	0.6756 ±0.0755	This paper
485	3WLGNN GNN baseline	0.6468 ±0.0588	This paper

ENZYMES

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GMN code ↗ Hyb global	0.7866	Memory-Based Graph Networks code ↗
2	MemGNN code ↗ Hyb global	0.7550	Memory-Based Graph Networks code ↗
3	TFGW SP Hyb global	0.7510 ±0.0500	Template based Graph Neural Network with Optimal Transport Distances
17	GatedGCN GNN baseline	0.7000 ±0.0494	This paper
20	GAT GNN baseline actual #20 vs paper #20	0.6850 ±0.0524	actual #20 vs paper #20 paper 0.6850 ±0.0524 prev best: Breaking the Limits of Message Passing Graph Neural Networks (ICML 2021) paper 0.6850 ±0.0524 Δ at pub 0.1440
26	GraphSAGE GNN baseline actual #26 vs paper #26	0.6817 ±0.0545	actual #26 vs paper #26 paper 0.6817 ±0.0545 prev best: Graph Neural Networks With Convolutional ARMA Filters (2019) Improvement within reported uncertainty; not highlighted. paper 0.6817 ±0.0545 Δ at pub 0.0787
32	GIN GNN baseline actual #32 vs paper #32	0.6767 ±0.0583	actual #32 vs paper #32 paper 0.6767 ±0.0583 prev best: Template based Graph Neural Network with Optimal Transport Distances (NeurIPS 2022) paper 0.6767 ±0.0583 Δ at pub 0.0547
36	GCN code ↗ GNN global	0.6650 ±0.0691	Hierarchical Graph Capsule Network code ↗
42	vanilla GCN GNN baseline	0.6583 ±0.0461	This paper
67	MoNet GNN baseline	0.6300 ±0.0809	This paper
85	3WLGNN GNN baseline	0.6100 ±0.0680	This paper
124	MLP Trad baseline actual #124 vs paper #124	0.5583 ±0.0352	actual #124 vs paper #124 paper 0.5583 ±0.0352 prev best: Breaking the Limits of Message Passing Graph Neural Networks (ICML 2021) paper 0.5583 ±0.0352 Δ at pub 0.2503
172	RingGNN GNN baseline	0.4533 ±0.0452	This paper

OGB

ogbl-collab

Global leaderboard →

#	Model	Hits@50 ↑	Paper
1	PROXI code ↗ Trad global	0.7650 ±0.0027	PROXI: Challenging the GNNs for Link Prediction code ↗
2	OGB Leader code ↗ global	0.7129 ±0.0018	PROXI: Challenging the GNNs for Link Prediction code ↗
3	GIDN GNN global	0.7096 ±0.0055	GIDN: A Lightweight Graph Inception Diffusion Network for High-efficient Link Prediction
9	GCN GNN baseline actual #9 vs paper #75 66 places worse	0.4896 ±0.0114	actual #9 vs paper #75 66 places worse paper 0.4896 ±0.0114 prev best: OGB (NeurIPS 2020) best today: On the Impact of Feature Heterophily on Link Prediction with Graph Neural Networks (NeurIPS 2024) Improvement within reported uncertainty; not highlighted. paper 0.4896 ±0.0114 Δ at pub 0.0182 Δ today 0.1725
28	SAGE code ↗ GNN global	0.5944 ±0.0137	PROXI: Challenging the GNNs for Link Prediction code ↗
34	GAT GNN baseline actual #34 vs paper #56 22 places worse	0.5150 ±0.0096	actual #34 vs paper #56 22 places worse paper 0.5150 ±0.0096 prev best: Neo-GNNs: Neighborhood Overlap-aware Graph Neural Networks for Link Prediction (NeurIPS 2022) best today: Asymmetric Learning for Graph Neural Network based Link Prediction (2023) paper 0.5150 ±0.0096 Δ at pub 0.0661 Δ today 0.0493
41	GraphSAGE GNN baseline actual #41 vs paper #56 15 places worse	0.5463 ±0.0112 best reported	actual #41 vs paper #56 15 places worse paper 0.5162 ±0.0069 best at pub: OGB (NeurIPS 2020) best today: Bring Your Own View: Graph Neural Networks for Link Prediction with Personalized Subgraph Selection (2022) paper 0.5162 ±0.0069 Δ at pub 0.0301 Δ today 0.0301
51	GatedGCN-PE GNN baseline	0.5285 ±0.0135	This paper
53	GatedGCN GNN baseline	0.5263 ±0.0117	This paper
59	vanilla GCN GNN baseline	0.5042 ±0.0113	This paper
71	GatedGCN-E GNN baseline	0.4921 ±0.0156	This paper
94	MLP Trad baseline actual #94 vs paper #110 16 places worse	0.2035 ±0.0217	actual #94 vs paper #110 16 places worse paper 0.2035 ±0.0217 prev best: Neo-GNNs: Neighborhood Overlap-aware Graph Neural Networks for Link Prediction (NeurIPS 2022) best today: Weak Models Can be Good Teachers: A Case Study on Link Prediction with MLPs (2025) Improvement within reported uncertainty; not highlighted. paper 0.2035 ±0.0217 Δ at pub 0.0037 Δ today 0.2403
95	Matrix Fact. Trad 60.5M baseline	0.4421 ±0.0045	This paper
99	GIN GNN baseline	0.4173 ±0.0228	This paper
105	MoNet GNN baseline	0.3614 ±0.0219	This paper

GNNBenchmark

MNIST

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	Schrödinger GNN global	0.9913 ±0.0004	Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables
2	CNN GNN global	0.9907 ±0.0007	Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables
3	MPNN GNN global	0.9895 ±0.0006	Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables
11	GatedGCN GNN baseline actual #11 vs paper #90 79 places worse	0.9855 ±0.0002 best reported	actual #11 vs paper #90 79 places worse paper 0.9734 ±0.0014 best at pub: Formatting Instructions for ICLR 2021 Conference Submissions (2020) best today: RandAlign: A Parameter-Free Method for Regularizing Graph Convolutional Networks (2024) paper 0.9734 ±0.0014 Δ at pub 0.0090 Δ today 0.0121
32	GIN GNN baseline actual #32 vs paper #110 78 places worse	0.9833 ±0.0011 best reported	actual #32 vs paper #110 78 places worse paper 0.9648 ±0.0025 best at pub: Graph-Coupled Oscillator Networks (ICML 2022) best today: Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables (2026) paper 0.9648 ±0.0025 Δ at pub 0.0075 Δ today 0.0184
86	GAT GNN baseline actual #86 vs paper #121 35 places worse	0.9755 ±0.0003 best reported	actual #86 vs paper #121 35 places worse paper 0.9554 ±0.0021 best at pub: Formatting Instructions for ICLR 2021 Conference Submissions (2020) best today: RandAlign: A Parameter-Free Method for Regularizing Graph Convolutional Networks (2024) paper 0.9554 ±0.0021 Δ at pub 0.0096 Δ today 0.0202
95	GraphSAGE GNN baseline	0.9731 ±0.0010	This paper
97	SAGE code ↗ GNN global	0.9731 ±0.0010	HyperAggregation: Aggregating over Graph Edges with Hypernetworks code ↗
126	MLP Trad baseline	0.9534 ±0.0014	This paper
129	3WLGNN GNN baseline	0.9507 ±0.0096	This paper
135	GCN GNN baseline actual #135 vs paper #141 6 places worse	0.9405 ±0.0005 best reported	actual #135 vs paper #141 6 places worse paper 0.9012 ±0.0014 best at pub: A Generalization of ViT/MLP-Mixer to Graphs (ICML 2022) best today: RandAlign: A Parameter-Free Method for Regularizing Graph Convolutional Networks (2024) paper 0.9012 ±0.0014 Δ at pub 0.0257 Δ today 0.0393
137	MoNet GNN baseline actual #137 vs paper #137	0.9111 best reported	actual #137 vs paper #137 paper 0.9081 ±0.0003 best at pub: Graph-Coupled Oscillator Networks (ICML 2022) paper 0.9081 ±0.0003 Δ at pub 0.0030
139	vanilla GCN GNN baseline	0.9071 ±0.0022	This paper
153	RingGNN GNN baseline	0.1135	This paper

CIFAR10

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	RandAlign GNN global	0.7942 ±0.0007	RandAlign: A Parameter-Free Method for Regularizing Graph Convolutional Networks
2	differential encoding global	0.7907 ±0.0027	Differential Encoding for Improved Representation Learning Over Graphs
3	PPGT GT global	0.7856 ±0.0070	Plain Transformers Can be Powerful Graph Learners
62	GatedGCN GNN baseline actual #62 vs paper #90 28 places worse	0.7133 ±0.0039 best reported	actual #62 vs paper #90 28 places worse paper 0.6731 ±0.0031 best at pub: Formatting Instructions for ICLR 2021 Conference Submissions (2020) paper 0.6731 ±0.0031 Δ at pub 0.0402
92	GraphSAGE GNN baseline actual #92 vs paper #104 12 places worse	0.6690 ±0.0020 best reported	actual #92 vs paper #104 12 places worse paper 0.6577 ±0.0031 best at pub: Beyond Low-pass Filtering: Graph Convolutional Networks with Automatic Filtering (2021) paper 0.6577 ±0.0031 Δ at pub 0.0113
101	GAT GNN baseline actual #101 vs paper #112 11 places worse	0.6611 ±0.0098 best reported	actual #101 vs paper #112 11 places worse paper 0.6422 ±0.0046 best at pub: Formatting Instructions for ICLR 2021 Conference Submissions (2020) paper 0.6422 ±0.0046 Δ at pub 0.0189
103	MoNet GNN baseline actual #103 vs paper #129 26 places worse	0.6591 ±0.0251 best reported	actual #103 vs paper #129 26 places worse paper 0.5465 ±0.0052 best at pub: SoGCN: Second-Order Graph Convolutional Networks (2021) paper 0.5465 ±0.0052 Δ at pub 0.1126
104	SAGE code ↗ GNN global	0.6577 ±0.0031	HyperAggregation: Aggregating over Graph Edges with Hypernetworks code ↗
109	GIN GNN baseline actual #109 vs paper #129 20 places worse	0.6510 ±0.0050 best reported	actual #109 vs paper #129 20 places worse paper 0.5525 ±0.0153 best at pub: Beyond Low-pass Filtering: Graph Convolutional Networks with Automatic Filtering (2021) paper 0.5525 ±0.0153 Δ at pub 0.0985
123	3WLGNN GNN baseline	0.5917 ±0.0159	This paper
124	GCN GNN baseline actual #124 vs paper #131 7 places worse	0.5839 ±0.0073 best reported	actual #124 vs paper #131 7 places worse paper 0.5414 ±0.0039 best at pub: From Stars to Subgraphs: Uplifting Any GNN with Local Structure Awareness (ICLR 2021) paper 0.5414 ±0.0039 Δ at pub 0.0425
126	MLP Trad baseline actual #126 vs paper #126	0.5634 ±0.0018	actual #126 vs paper #126 paper 0.5634 ±0.0018 prev best: Beyond Low-pass Filtering: Graph Convolutional Networks with Automatic Filtering (2021) paper 0.5634 ±0.0018 Δ at pub 0.0274
129	vanilla GCN GNN baseline	0.5571 ±0.0038	This paper
140	RingGNN GNN baseline	0.1930 ±0.1611	This paper

CLUSTER

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GRPE-Small Hyb 489.0M global	0.8159 ±0.0019	GRPE: Relative Positional Encoding for Graph Transformer
2	TANGO-GPS Hyb global	0.8011 ±0.0014	TANGO: Graph Neural Dynamics via Learned Energy and Tangential Flows
3	GRIT GNN global	0.8003 ±0.0028	GCCM: Enhancing Generative Graph Prediction via Contrastive Consistency Model
55	GatedGCN-PE ext data GNN baseline	0.7608 ±0.0020	This paper
54	GatedGCN GNN 505.0M global	0.7608 ±0.0020	GRPE: Relative Positional Encoding for Graph Transformer
67	GAT GNN baseline actual #67 vs paper #90 23 places worse	0.7059 ±0.0045 best reported	actual #67 vs paper #90 23 places worse paper 0.5773 ±0.0032 best at pub: How Expressive are Transformers in Spectral Domain for Graphs? (TMLR 2022) best today: GCCM: Enhancing Generative Graph Prediction via Contrastive Consistency Model (2026) paper 0.5773 ±0.0032 Δ at pub 0.1286 Δ today 0.1286
70	GCN GNN baseline actual #70 vs paper #102 32 places worse	0.6903 ±0.0137 best reported	actual #70 vs paper #102 32 places worse paper 0.4783 ±0.0151 best at pub: How Expressive are Transformers in Spectral Domain for Graphs? (TMLR 2022) best today: Recurrent Distance Filtering for Graph Representation Learning (ICML 2023) paper 0.4783 ±0.0151 Δ at pub 0.2067 Δ today 0.2120
76	MoNet GNN baseline actual #76 vs paper #88 12 places worse	0.6641 ±0.0054 best reported	actual #76 vs paper #88 12 places worse paper 0.5806 ±0.0013 best at pub: SoGCN: Second-Order Graph Convolutional Networks (2021) best today: RandAlign: A Parameter-Free Method for Regularizing Graph Convolutional Networks (2024) paper 0.5806 ±0.0013 Δ today 0.0834
79	GIN GNN baseline actual #79 vs paper #87 8 places worse	0.6472 ±0.0155 best reported	actual #79 vs paper #87 8 places worse paper 0.5838 ±0.0024 best at pub: How Expressive are Transformers in Spectral Domain for Graphs? (TMLR 2022) best today: GCCM: Enhancing Generative Graph Prediction via Contrastive Consistency Model (2026) paper 0.5838 ±0.0024 Δ at pub 0.0633 Δ today 0.0633
80	GraphSAGE GNN baseline actual #80 vs paper #94 14 places worse	0.6388 ±0.0011 best reported	actual #80 vs paper #94 14 places worse paper 0.5045 ±0.0014 best at pub: SoGCN: Second-Order Graph Convolutional Networks (2021) best today: IJCAI--23 Formatting Instructions (IJCAI 2023) paper 0.5045 ±0.0014 Δ today 0.1343
85	3WLGNN GNN baseline actual #85 vs paper #90 5 places worse	0.6040 ±0.0042 best reported	actual #85 vs paper #90 5 places worse paper 0.5713 ±0.0654 best at pub: SoGCN: Second-Order Graph Convolutional Networks (2021) Difference within reported uncertainty; not counted as underperformance. paper 0.5713 ±0.0654 Δ at pub 0.0327
93	vanilla GCN GNN baseline	0.5344 ±0.0203	This paper
104	RingGNN GNN baseline	0.4242 ±0.2006	This paper
114	MLP Trad baseline	0.2097 ±0.0000	This paper

PATTERN

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	DeepGraph (48) code ↗ Hyb global	0.9066 ±0.0006	Are More Layers Beneficial to Graph Transformers? code ↗
2	GRIT (our run) GT global	0.9041 ±0.0023	Enhancing Graph Neural Networks with Quantum Computed Encodings
3	GRIT 2-QiQRW Hyb global	0.9016 ±0.0045	Enhancing Graph Neural Networks with Quantum Computed Encodings
67	GatedGCN-PE ext data GNN baseline	0.8651 ±0.0009	This paper
68	GatedGCN (Rand sign(EigVecs)) ext data baseline	0.8651 ±0.0009	This paper
66	GatedGCN code ↗ GNN 0.5M global	0.8651 ±0.0008	IJCAI--23 Formatting Instructions code ↗
69	GatedGCN (Abs(EigVecs)) ext data baseline	0.8639 ±0.0004	This paper
72	RingGNN GNN baseline	0.8625 ±0.0001	This paper
74	GatedGCN (EigVecs-2) ext data GNN baseline	0.8603 ±0.0009	This paper
78	GatedGCN (Rand node ordering) GNN baseline	0.8577 ±0.0004	This paper
85	3WLGNN GNN baseline	0.8566 ±0.0035	This paper
88	GCN GNN baseline actual #88 vs paper #92 4 places worse	0.8550 ±0.0004	actual #88 vs paper #92 4 places worse paper 0.8550 ±0.0004 prev best: How Expressive are Transformers in Spectral Domain for Graphs? (TMLR 2022) best today: Recurrent Distance Filtering for Graph Representation Learning (ICML 2023) paper 0.8550 ±0.0004 Δ at pub 0.1361 Δ today 0.0012
89	GIN GNN baseline	0.8559 ±0.0001	This paper
91	MoNet GNN baseline actual #91 vs paper #92 1 place worse	0.8558 ±0.0004 best reported	actual #91 vs paper #92 1 place worse paper 0.8548 ±0.0004 best at pub: SoGCN: Second-Order Graph Convolutional Networks (2021) best today: RandAlign: A Parameter-Free Method for Regularizing Graph Convolutional Networks (2024) paper 0.8548 ±0.0004 Δ today 0.0010
107	GAT GNN baseline actual #107 vs paper #112 5 places worse	0.7827 ±0.0019 best reported	actual #107 vs paper #112 5 places worse paper 0.7582 ±0.0182 best at pub: How Expressive are Transformers in Spectral Domain for Graphs? (TMLR 2022) best today: GCCM: Enhancing Generative Graph Prediction via Contrastive Consistency Model (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.7582 ±0.0182 Δ at pub 0.0245 Δ today 0.0245
119	vanilla GCN GNN baseline	0.6388 ±0.0007	This paper
123	MLP Trad baseline	0.5052	This paper
124	GraphSAGE GNN baseline	0.5052 ±0.0000	This paper

CSL

Global leaderboard →

#	Model	Accuracy ↑	Paper
8	GCN GNN baseline actual #8 vs paper #8	1.0000	actual #8 vs paper #8 paper 1.0000 prev best: A Generalization of ViT/MLP-Mixer to Graphs (ICML 2022) paper 1.0000 Δ at pub 1.0000
1	PPGN code ↗ GNN global	1.0000	Revisiting Random Walks for Learning on Graphs code ↗
2	RW-AgentNet code ↗ Walk global	1.0000	Revisiting Random Walks for Learning on Graphs code ↗
3	AgentNet code ↗ Trad global	1.0000	Revisiting Random Walks for Learning on Graphs code ↗
33	MoNet GNN baseline	0.9997 ±0.0033	This paper
35	GraphSAGE GNN baseline	0.9993 ±0.0047	This paper
34	GAT GNN baseline	0.9993 ±0.0047	This paper
36	GatedGCN GNN baseline actual #36 vs paper #36	0.9960 ±0.0108	actual #36 vs paper #36 paper 0.9960 ±0.0108 prev best: A Generalization of ViT/MLP-Mixer to Graphs (ICML 2022) paper 0.9960 ±0.0108 Δ at pub 0.9960
37	GatedGCN (Rand sign(EigVecs)) baseline	0.9960 ±0.0108	This paper
38	GatedGCN (Abs(EigVecs)) baseline	0.9943 ±0.0113	This paper
40	GIN GNN baseline actual #40 vs paper #40	0.9933 ±0.0133	actual #40 vs paper #40 paper 0.9933 ±0.0133 prev best: On the Equivalence between Graph Isomorphism Testing and Function Approximation with GNNs (NeurIPS 2019) paper 0.9933 ±0.0133 Δ at pub 0.8933
46	3WLGNN GNN baseline actual #46 vs paper #47 1 place worse	0.9780 ±0.1090 best reported	actual #46 vs paper #47 1 place worse paper 0.9570 ±0.1485 best at pub: Weisfeiler and Lehman Go Cellular: CW Networks (NeurIPS 2021) best today: Graph Self-Supervised Learning with Learnable Structural and Positional Encodings (WWW 2025) Difference within reported uncertainty; not counted as underperformance. paper 0.9570 ±0.1485 Δ at pub 0.0210 Δ today 0.0210
69	RingGNN GNN baseline actual #69 vs paper #69	0.1723 ±0.0633	actual #69 vs paper #69 paper 0.1723 ±0.0633 prev best: Weisfeiler and Lehman Go Cellular: CW Networks (NeurIPS 2021) paper 0.1723 ±0.0633 Δ at pub 0.0723
79	All MP-GCNs GNN baseline	0.1000	This paper

TSP

Global leaderboard →

#	Model	F1 ↑	Paper
1	TGT-Agx4 code ↗ GT global	0.8710 ±0.0010	Triplet Interaction Improves Graph Transformers: Accurate Molecular Graph Learning with Triplet … code ↗
2	TGT-Agx3 code ↗ GT global	0.8660 ±0.0010	Triplet Interaction Improves Graph Transformers: Accurate Molecular Graph Learning with Triplet … code ↗
3	GIG-GatedGCN code ↗ Hyb global	0.8630 ±0.0010	Graph in Graph Neural Network code ↗
8	GatedGCN GNN baseline actual #8 vs paper #11 3 places worse	0.8380 ±0.0020 best reported	actual #8 vs paper #11 3 places worse paper 0.7910 ±0.0030 best at pub: Formatting Instructions for ICLR 2021 Conference Submissions (2020) best today: Graph in Graph Neural Network (2024) paper 0.7910 ±0.0030 Δ at pub 0.0170 Δ today 0.0470
9	GatedGCN-E GNN baseline	0.8380 ±0.0020	This paper
12	RingGNN GNN baseline	0.7040 ±0.0030	This paper
15	3WLGNN GNN baseline	0.6940 ±0.0730	This paper
16	k-NN Heuristic Trad baseline	0.6930	This paper
18	GAT GNN baseline	0.6710 ±0.0020	This paper
19	GraphSAGE GNN baseline	0.6650 ±0.0030	This paper
20	GIN GNN baseline	0.6560 ±0.0030	This paper
21	GCN GNN baseline actual #21 vs paper #21	0.6430 ±0.0010	actual #21 vs paper #21 paper 0.6430 ±0.0010 prev best: Formatting Instructions for ICLR 2021 Conference Submissions (2020) paper 0.6430 ±0.0010 Δ at pub 0.0130
22	MoNet GNN baseline	0.6410 ±0.0020	This paper
23	vanilla GCN GNN baseline	0.6300 ±0.0010	This paper
24	MLP Trad baseline	0.5440 ±0.0010	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 Improved at pub — later paper surpassed this result Best today — OK at publication; today's leaderboard best is elsewhere