Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification

Brown Zaz, M. Catala, Ferran Hernandez Caralt, Moshe Eliasof, Pietro Liò

Abstract

In the transductive setting, where the full graph is observed but node labels are only partially available, progress in semi-supervised node classification has largely focused on architectural innovation. In this paper, we revisit an orthogonal axis: the training objective. We start from a simple observation: transductive models produce predictions for every node during training, including nodes without labels. These unlabeled-node predictions may contain useful training signal, but standard supervised objectives discard them because no ground-truth labels are available. Inspired by the decomposition of cross-entropy into a label-dependent alignment term and a label-independent entropy term, we propose prediction confidence as a natural way to extract this signal in the absence of labels. This motivates Transductive Sharpening (TS): a loss-level modification that minimizes prediction entropy on unlabeled nodes while counterbalancing this effect on labeled nodes. We evaluate Transductive Sharpening across a wide range of node-classification benchmarks and observe consistent performance improvements without requiring any changes to the backbone architecture. Code is available at.

Benchmark Results

Tested on 11 datasets 60 baseline values reported too low Non-standard metric on 1 dataset

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 baseline actual #21 vs paper #220 199 places worse	0.9004 best reported	actual #21 vs paper #220 199 places worse paper 0.8516 ±0.0036 best at pub: Understanding Heterophily for Graph Neural Networks (ICML 2024) paper 0.8516 ±0.0036 Δ at pub 0.0488
39	SAGE GNN baseline actual #39 vs paper #228 189 places worse	0.8911 ±0.0012 best reported	actual #39 vs paper #228 189 places worse paper 0.8494 ±0.0089 best at pub: Efficient Text-Attributed Graph Learning through Selective Annotation and Graph Alignment (TMLR 2025) paper 0.8494 ±0.0089 Δ at pub 0.0417
44	GAT GNN baseline actual #44 vs paper #325 281 places worse	0.8897 ±0.0014 best reported	actual #44 vs paper #325 281 places worse paper 0.8370 ±0.0119 best at pub: Adaptive Tokenization: On the Hop-Overpriority Problem in Tokenized Graph Learning Models (2025) paper 0.8370 ±0.0119 Δ at pub 0.0527
68	NodeFormer GT baseline actual #68 vs paper #447 379 places worse	0.8848 ±0.0033 best reported	actual #68 vs paper #447 379 places worse paper 0.8273 ±0.0075 best at pub: Efficient Tuning and Inference for Large Language Models on Textual Graphs (IJCAI 2024) paper 0.8273 ±0.0075 Δ at pub 0.0575
141	GraphSAGE GNN global	0.8708 ±0.0085	Large Language Model Meets Graph Neural Network in Knowledge Distillation
175	OGFormer GT baseline	0.8640 ±0.0030	This paper
186	NormProp GNN baseline actual #186 vs paper #205 19 places worse	0.8607 best reported	actual #186 vs paper #205 19 places worse paper 0.8546 ±0.0051 best at pub: Rethinking Semi-Supervised Node Classification with Self-Supervised Graph Clustering (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.8546 ±0.0051 Δ at pub 0.0061
196	GCN+TS GNN baseline actual #196 vs paper #196	0.8574 ±0.0054	actual #196 vs paper #196 paper 0.8574 ±0.0054 prev best: TSC: A Simple Two-Sided Constraint against Over-Smoothing (KDD 2024) Improvement within reported uncertainty; not highlighted. paper 0.8574 ±0.0054 Δ at pub 0.0064
214	SAGE+T GNN baseline	0.8528 ±0.0111	This paper
233	SGFormer GT baseline actual #233 vs paper #233	0.8482 ±0.0085	actual #233 vs paper #233 paper 0.8482 ±0.0085 prev best: Latent-Hysteresis Graph ODEs: Modeling Coupled Topology-Feature Evolution via Continuous Phase Transitions (2026) Improvement within reported uncertainty; not highlighted. paper 0.8482 ±0.0085 Δ at pub 0.0023
244	GAT+T Hyb baseline	0.8462 ±0.0089	This paper
277	ELU-GCN Hyb baseline	0.8429 ±0.0039	This paper
291	Exphormer GT baseline actual #291 vs paper #400 109 places worse	0.8410 ±0.0080 best reported	actual #291 vs paper #400 109 places worse paper 0.8329 ±0.0136 best at pub: Diffusion on Graph: Augmentation of Graph Structure for Node Classification (TMLR 2025) Difference within reported uncertainty; not counted as underperformance. paper 0.8329 ±0.0136 Δ at pub 0.0081
314	GraphGPS Hyb baseline actual #314 vs paper #314	0.8387 ±0.0096	actual #314 vs paper #314 paper 0.8387 ±0.0096 prev best: Latent-Hysteresis Graph ODEs: Modeling Coupled Topology-Feature Evolution via Continuous Phase Transitions (2026) Improvement within reported uncertainty; not highlighted. paper 0.8387 ±0.0096 Δ at pub 0.0092
381	Polynormer GT baseline actual #381 vs paper #381	0.8343 ±0.0089	actual #381 vs paper #381 paper 0.8343 ±0.0089 prev best: Latent-Hysteresis Graph ODEs: Modeling Coupled Topology-Feature Evolution via Continuous Phase Transitions (2026) Improvement within reported uncertainty; not highlighted. paper 0.8343 ±0.0089 Δ at pub 0.0019
401	GOAT GT baseline	0.8326 ±0.0124	This paper
516	NAGphormer GT baseline actual #516 vs paper #612 96 places worse	0.8213 ±0.0118 best reported	actual #516 vs paper #612 96 places worse paper 0.8092 ±0.0117 best at pub: Toward a Unified Geometry Understanding: Riemannian Diffusion Framework for Graph Generation and Prediction (NeurIPS 2025) Difference within reported uncertainty; not counted as underperformance. paper 0.8092 ±0.0117 Δ at pub 0.0121
724	MLP Trad baseline actual #724 vs paper #872 148 places worse	0.7768 best reported	actual #724 vs paper #872 148 places worse paper 0.6096 ±0.0251 best at pub: Understanding Heterophily for Graph Neural Networks (ICML 2024) paper 0.6096 ±0.0251 Δ at pub 0.1672
866	MLP+TS Trad baseline	0.6448 ±0.0283	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 ↗
7	NodeFormer GT baseline actual #7 vs paper #500 493 places worse	0.9490 ±0.0019 best reported	actual #7 vs paper #500 493 places worse paper 0.7959 ±0.0092 best at pub: Adaptive Tokenization: On the Hop-Overpriority Problem in Tokenized Graph Learning Models (2025) paper 0.7959 ±0.0092 Δ at pub 0.1531
18	GCN GNN baseline actual #18 vs paper #380 362 places worse	0.9296 ±0.0015 best reported	actual #18 vs paper #380 362 places worse paper 0.8088 ±0.0031 best at pub: Adaptive Tokenization: On the Hop-Overpriority Problem in Tokenized Graph Learning Models (2025) paper 0.8088 ±0.0031 Δ at pub 0.1208
20	GAT GNN baseline actual #20 vs paper #474 454 places worse	0.9233 ±0.0018 best reported	actual #20 vs paper #474 454 places worse paper 0.7980 ±0.0093 best at pub: Adaptive Tokenization: On the Hop-Overpriority Problem in Tokenized Graph Learning Models (2025) paper 0.7980 ±0.0093 Δ at pub 0.1253
53	Polynormer GT baseline actual #53 vs paper #541 488 places worse	0.9044 ±0.0021 best reported	actual #53 vs paper #541 488 places worse paper 0.7935 ±0.0073 best at pub: VecFormer: Towards Efficient and Generalizable Graph Transformer with Graph Token Attention (WWW 2026) paper 0.7935 ±0.0073 Δ at pub 0.1109
56	SGFormer GT baseline actual #56 vs paper #398 342 places worse	0.9037 ±0.0020 best reported	actual #56 vs paper #398 342 places worse paper 0.8060 ±0.0049 best at pub: VecFormer: Towards Efficient and Generalizable Graph Transformer with Graph Token Attention (WWW 2026) paper 0.8060 ±0.0049 Δ at pub 0.0977
91	NAGphormer GT baseline actual #91 vs paper #437 346 places worse	0.8939 ±0.0013 best reported	actual #91 vs paper #437 346 places worse paper 0.8014 ±0.0106 best at pub: NTFormer: A Composite Node Tokenized Graph Transformer for Node Classification (2024) paper 0.8014 ±0.0106 Δ at pub 0.0925
99	GOAT GT baseline actual #99 vs paper #445 346 places worse	0.8915 ±0.0039 best reported	actual #99 vs paper #445 346 places worse paper 0.8006 ±0.0067 best at pub: VecFormer: Towards Efficient and Generalizable Graph Transformer with Graph Token Attention (WWW 2026) paper 0.8006 ±0.0067 Δ at pub 0.0909
105	Exphormer GT baseline actual #105 vs paper #484 379 places worse	0.8906 ±0.0024 best reported	actual #105 vs paper #484 379 places worse paper 0.7967 ±0.0073 best at pub: VecFormer: Towards Efficient and Generalizable Graph Transformer with Graph Token Attention (WWW 2026) paper 0.7967 ±0.0073 Δ at pub 0.0939
116	SAGE GNN baseline actual #116 vs paper #646 530 places worse	0.8881 ±0.0015 best reported	actual #116 vs paper #646 530 places worse paper 0.7840 ±0.0112 best at pub: Efficient Text-Attributed Graph Learning through Selective Annotation and Graph Alignment (TMLR 2025) paper 0.7840 ±0.0112 Δ at pub 0.1041
125	GraphSAGE code ↗ GNN global	0.8864 ±0.0030	Non-backtracking Graph Neural Networks code ↗
172	GraphGPS Hyb baseline actual #172 vs paper #458 286 places worse	0.8772 ±0.0030 best reported	actual #172 vs paper #458 286 places worse paper 0.7994 ±0.0026 best at pub: VecFormer: Towards Efficient and Generalizable Graph Transformer with Graph Token Attention (WWW 2026) paper 0.7994 ±0.0026 Δ at pub 0.0778
210	MLP Trad baseline actual #210 vs paper #846 636 places worse	0.8690 ±0.0074 best reported	actual #210 vs paper #846 636 places worse paper 0.6896 ±0.0116 best at pub: On the Equivalence of Graph Convolution and Mixup (TMLR 2023) paper 0.6896 ±0.0116 Δ at pub 0.1794
343	NormProp GNN baseline actual #343 vs paper #393 50 places worse	0.8153 ±0.0001 best reported	actual #343 vs paper #393 50 places worse paper 0.8072 ±0.0109 best at pub: Rethinking Semi-Supervised Node Classification with Self-Supervised Graph Clustering (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.8072 ±0.0109 Δ at pub 0.0081
348	OGFormer GT baseline	0.8150 ±0.0050	This paper
392	GCN+TS GNN baseline actual #392 vs paper #392	0.8074 ±0.0030	actual #392 vs paper #392 paper 0.8074 ±0.0030 prev best: TSC: A Simple Two-Sided Constraint against Over-Smoothing (KDD 2024) Improvement within reported uncertainty; not highlighted. paper 0.8074 ±0.0030 Δ at pub 0.0004
403	ELU-GCN Hyb baseline	0.8051 ±0.0021	This paper
482	SAGE+T GNN baseline	0.7972 ±0.0069	This paper
611	GAT+T Hyb baseline	0.7884 ±0.0067	This paper
822	MLP+TS Trad baseline	0.7230 ±0.0189	This paper

Amazon-Photo

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GraphTARIF Hyb baseline	0.9703 ±0.0019	This paper
2	GAT GNN baseline actual #2 vs paper #2	0.9669 ±0.0014	actual #2 vs paper #2 paper 0.9669 ±0.0014 prev best: SignGT: Signed Attention-based Graph Transformer for Graph Representation Learning (2023) paper 0.9669 ±0.0014 Δ at pub 0.0171
3	GAT+T Hyb baseline	0.9660 ±0.0007	This paper
4	Polynormer GT baseline actual #4 vs paper #4	0.9657 ±0.0023	actual #4 vs paper #4 paper 0.9657 ±0.0023 prev best: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) paper 0.9657 ±0.0023 Δ at pub 0.0046
5	SAGE+T GNN baseline	0.9651 ±0.0023	This paper
7	SAGE GNN baseline actual #7 vs paper #7	0.9643 ±0.0027	actual #7 vs paper #7 paper 0.9643 ±0.0027 prev best: VecFormer: Towards Efficient and Generalizable Graph Transformer with Graph Token Attention (WWW 2026) paper 0.9643 ±0.0027 Δ at pub 0.0240
8	GCN GNN baseline actual #8 vs paper #8	0.9632 ±0.0008	actual #8 vs paper #8 paper 0.9632 ±0.0008 prev best: SignGT: Signed Attention-based Graph Transformer for Graph Representation Learning (2023) paper 0.9632 ±0.0008 Δ at pub 0.0156
9	GCN+TS GNN baseline	0.9621 ±0.0011	This paper
11	NAGphormer GT baseline actual #11 vs paper #11	0.9614 ±0.0016	actual #11 vs paper #11 paper 0.9614 ±0.0016 prev best: NTFormer: A Composite Node Tokenized Graph Transformer for Node Classification (2024) Improvement within reported uncertainty; not highlighted. paper 0.9614 ±0.0016 Δ at pub 0.0020
23	Exphormer GT baseline actual #23 vs paper #23	0.9569 ±0.0039	actual #23 vs paper #23 paper 0.9569 ±0.0039 prev best: Even Sparser Graph Transformers (NeurIPS 2024) Improvement within reported uncertainty; not highlighted. paper 0.9569 ±0.0039 Δ at pub 0.0033
29	SGFormer GT baseline actual #29 vs paper #29	0.9558 ±0.0036	actual #29 vs paper #29 paper 0.9558 ±0.0036 prev best: VecFormer: Towards Efficient and Generalizable Graph Transformer with Graph Token Attention (WWW 2026) paper 0.9558 ±0.0036 Δ at pub 0.0077
32	OGFormer GT baseline	0.9550	This paper
44	NodeFormer GT baseline actual #44 vs paper #158 114 places worse	0.9527 ±0.0022 best reported	actual #44 vs paper #158 114 places worse paper 0.9343 ±0.0056 best at pub: SignGT: Signed Attention-based Graph Transformer for Graph Representation Learning (2023) best today: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) paper 0.9343 ±0.0056 Δ at pub 0.0184 Δ today 0.0184
56	GraphSAGE code ↗ GNN global	0.9508 ±0.0026	A Quasi-Wasserstein Loss for Learning Graph Neural Networks code ↗
57	GraphGPS Hyb baseline actual #57 vs paper #65 8 places worse	0.9506 ±0.0013 best reported	actual #57 vs paper #65 8 places worse paper 0.9489 ±0.0014 best at pub: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) best today: Can TabPFN Compete with GNNs for Node Classification via Graph Tabularization? (LoG 2025) Difference within reported uncertainty; not counted as underperformance. paper 0.9489 ±0.0014 Δ at pub 0.0017 Δ today 0.0017
86	GOAT GT baseline actual #86 vs paper #86	0.9433 ±0.0021	actual #86 vs paper #86 paper 0.9433 ±0.0021 prev best: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) Improvement within reported uncertainty; not highlighted. paper 0.9433 ±0.0021 Δ at pub 0.0137
390	MLP Trad baseline actual #390 vs paper #494 104 places worse	0.9128 ±0.0063 best reported	actual #390 vs paper #494 104 places worse paper 0.8757 ±0.0052 best at pub: Clarify Confused Nodes via Separated Learning (2023) paper 0.8757 ±0.0052 Δ at pub 0.0371
492	MLP+TS Trad baseline	0.8765 ±0.0022	This paper

Amazon-Computers

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GraphTARIF Hyb baseline	0.9461 ±0.0017	This paper
2	GCN GNN baseline actual #2 vs paper #2	0.9412 ±0.0008	actual #2 vs paper #2 paper 0.9412 ±0.0008 prev best: Diss-l-ECT: Dissecting Graph Data with local Euler Characteristic Transforms (ICML 2024) Improvement within reported uncertainty; not highlighted. paper 0.9412 ±0.0008 Δ at pub 0.0252
3	GAT GNN baseline actual #3 vs paper #3	0.9398 ±0.0022	actual #3 vs paper #3 paper 0.9398 ±0.0022 prev best: Diss-l-ECT: Dissecting Graph Data with local Euler Characteristic Transforms (ICML 2024) Improvement within reported uncertainty; not highlighted. paper 0.9398 ±0.0022 Δ at pub 0.0158
4	GCN+TS GNN baseline	0.9398 ±0.0026	This paper
5	GAT+T Hyb baseline	0.9386 ±0.0013	This paper
6	Polynormer GT baseline actual #6 vs paper #6	0.9378 ±0.0010	actual #6 vs paper #6 paper 0.9378 ±0.0010 prev best: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) paper 0.9378 ±0.0010 Δ at pub 0.0060
8	SAGE GNN baseline actual #8 vs paper #8	0.9351 ±0.0006	actual #8 vs paper #8 paper 0.9351 ±0.0006 prev best: Automatic Graph Topology-Aware Transformer (2024) paper 0.9351 ±0.0006 Δ at pub 0.0299
9	SAGE+T GNN baseline	0.9343 ±0.0011	This paper
12	OGFormer GT baseline	0.9290 ±0.0030	This paper
18	SGFormer GT baseline actual #18 vs paper #18	0.9242 ±0.0066	actual #18 vs paper #18 paper 0.9242 ±0.0066 prev best: VecFormer: Towards Efficient and Generalizable Graph Transformer with Graph Token Attention (WWW 2026) paper 0.9242 ±0.0066 Δ at pub 0.0167
20	GOAT GT baseline actual #20 vs paper #20	0.9229 ±0.0037	actual #20 vs paper #20 paper 0.9229 ±0.0037 prev best: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) paper 0.9229 ±0.0037 Δ at pub 0.0133
33	Exphormer GT baseline actual #33 vs paper #33	0.9180 ±0.0035	actual #33 vs paper #33 paper 0.9180 ±0.0035 prev best: Revisiting the Necessity of Graph Learning and Common Graph Benchmarks (2024) Improvement within reported uncertainty; not highlighted. paper 0.9180 ±0.0035 Δ at pub 0.0020
34	GraphGPS Hyb baseline actual #34 vs paper #34	0.9179 ±0.0063	actual #34 vs paper #34 paper 0.9179 ±0.0063 prev best: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) Improvement within reported uncertainty; not highlighted. paper 0.9179 ±0.0063 Δ at pub 0.0060
38	NAGphormer GT baseline actual #38 vs paper #38	0.9169 ±0.0030	actual #38 vs paper #38 paper 0.9169 ±0.0030 prev best: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) paper 0.9169 ±0.0030 Δ at pub 0.0047
46	GraphSAGE GNN global	0.9120 ±0.0029	Can TabPFN Compete with GNNs for Node Classification via Graph Tabularization?
48	NodeFormer GT baseline actual #48 vs paper #244 196 places worse	0.9112 ±0.0043 best reported	actual #48 vs paper #244 196 places worse paper 0.8729 ±0.0058 best at pub: SignGT: Signed Attention-based Graph Transformer for Graph Representation Learning (2023) paper 0.8729 ±0.0058 Δ at pub 0.0383
320	MLP Trad baseline actual #320 vs paper #390 70 places worse	0.8463 best reported	actual #320 vs paper #390 70 places worse paper 0.8246 ±0.0045 best at pub: GraphFM: A generalist graph transformer that learns transferable representations across diverse domains (TMLR 2024) paper 0.8246 ±0.0045 Δ at pub 0.0217
387	MLP+TS Trad baseline	0.8287 ±0.0056	This paper

Coauthor-Physics

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GrokFormer code ↗ GT global	0.9831 ±0.0018	GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers code ↗
2	PolyFormer code ↗ GT global	0.9808 ±0.0027	GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers code ↗
3	SGFormer GT baseline actual #3 vs paper #35 32 places worse	0.9796 ±0.0081 best reported	actual #3 vs paper #35 32 places worse paper 0.9675 ±0.0026 best at pub: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.9675 ±0.0026 Δ at pub 0.0121
5	NAGphormer GT baseline actual #5 vs paper #20 15 places worse	0.9785 ±0.0026 best reported	actual #5 vs paper #20 15 places worse paper 0.9735 ±0.0012 best at pub: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) paper 0.9735 ±0.0012 Δ at pub 0.0050
6	GAT GNN baseline actual #6 vs paper #20 14 places worse	0.9782 ±0.0028 best reported	actual #6 vs paper #20 14 places worse paper 0.9732 ±0.0006 best at pub: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) paper 0.9732 ±0.0006 Δ at pub 0.0050
9	GCN GNN baseline actual #9 vs paper #15 6 places worse	0.9774 ±0.0035 best reported	actual #9 vs paper #15 6 places worse paper 0.9744 ±0.0014 best at pub: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.9744 ±0.0014 Δ at pub 0.0030
13	GraphGPS Hyb baseline actual #13 vs paper #36 23 places worse	0.9760 ±0.0005 best reported	actual #13 vs paper #36 23 places worse paper 0.9671 ±0.0015 best at pub: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) paper 0.9671 ±0.0015 Δ at pub 0.0089
16	GCN+TS GNN baseline	0.9744 ±0.0014	This paper
17	SAGE GNN baseline actual #17 vs paper #23 6 places worse	0.9739 ±0.0018 best reported	actual #17 vs paper #23 6 places worse paper 0.9725 ±0.0008 best at pub: Clarify Confused Nodes via Separated Learning (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.9725 ±0.0008 Δ at pub 0.0014
18	GraphTARIF Hyb baseline	0.9739 ±0.0007	This paper
23	GAT+T Hyb baseline	0.9726 ±0.0003	This paper
25	SAGE+T GNN baseline	0.9723	This paper
26	Polynormer GT baseline actual #26 vs paper #27 1 place worse	0.9722 ±0.0006 best reported	actual #26 vs paper #27 1 place worse paper 0.9718 ±0.0011 best at pub: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.9718 ±0.0011 Δ at pub 0.0004
27	Exphormer GT baseline actual #27 vs paper #31 4 places worse	0.9720 ±0.0010 best reported	actual #27 vs paper #31 4 places worse paper 0.9706 ±0.0013 best at pub: Revisiting the Necessity of Graph Learning and Common Graph Benchmarks (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.9706 ±0.0013 Δ at pub 0.0014
34	NodeFormer GT baseline actual #34 vs paper #43 9 places worse	0.9680 ±0.0063 best reported	actual #34 vs paper #43 9 places worse paper 0.9648 ±0.0034 best at pub: VecFormer: Towards Efficient and Generalizable Graph Transformer with Graph Token Attention (WWW 2026) Difference within reported uncertainty; not counted as underperformance. paper 0.9648 ±0.0034 Δ at pub 0.0032
43	GraphSAGE code ↗ GNN global	0.9649 ±0.0006	Polynormer: Polynomial-Expressive Graph Transformer in Linear Time code ↗
44	GOAT GT baseline actual #44 vs paper #44	0.9647 ±0.0016	actual #44 vs paper #44 paper 0.9647 ±0.0016 prev best: VecFormer: Towards Efficient and Generalizable Graph Transformer with Graph Token Attention (WWW 2026) Improvement within reported uncertainty; not highlighted. paper 0.9647 ±0.0016 Δ at pub 0.0020
57	MLP+TS Trad baseline	0.9598 ±0.0008	This paper
58	MLP Trad baseline actual #58 vs paper #58	0.9597 ±0.0007	actual #58 vs paper #58 paper 0.9597 ±0.0007 prev best: Clarify Confused Nodes via Separated Learning (2023) Improvement within reported uncertainty; not highlighted. paper 0.9597 ±0.0007 Δ at pub 0.0034

WikiCS

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GCN GNN baseline actual #1 vs paper #91 90 places worse	0.9302 ±0.0011 best reported	actual #1 vs paper #91 90 places worse paper 0.8017 ±0.0048 best at pub: Augmentation-Free Graph Contrastive Learning with Performance Guarantee (2022) paper 0.8017 ±0.0048 Δ at pub 0.1285
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
5	SAGE+T GNN baseline actual #5 vs paper #73 68 places worse	0.8823 best reported	actual #5 vs paper #73 68 places worse paper 0.8110 ±0.0032 best at pub: Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning (2025) paper 0.8110 ±0.0032 Δ at pub 0.0713
9	SAGE GNN baseline actual #9 vs paper #78 69 places worse	0.8771 best reported	actual #9 vs paper #78 69 places worse paper 0.8077 ±0.0014 best at pub: Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning (2025) paper 0.8077 ±0.0014 Δ at pub 0.0694
31	GAT GNN baseline actual #31 vs paper #73 42 places worse	0.8533 best reported	actual #31 vs paper #73 42 places worse paper 0.8103 ±0.0092 best at pub: Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning (2025) paper 0.8103 ±0.0092 Δ at pub 0.0430
48	NAGphormer GT baseline actual #48 vs paper #176 128 places worse	0.8341 ±0.0034 best reported	actual #48 vs paper #176 128 places worse paper 0.7792 ±0.0093 best at pub: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) paper 0.7792 ±0.0093 Δ at pub 0.0549
54	GIN GNN global	0.8303 ±0.0021	Unleashing the Potential of Text-attributed Graphs: Automatic Relation Decomposition via Large …
57	SGFormer GT baseline actual #57 vs paper #97 40 places worse	0.8271 ±0.0056 best reported	actual #57 vs paper #97 40 places worse paper 0.8005 ±0.0046 best at pub: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) paper 0.8005 ±0.0046 Δ at pub 0.0266
74	NodeFormer GT baseline actual #74 vs paper #230 156 places worse	0.8103 ±0.0094 best reported	actual #74 vs paper #230 156 places worse paper 0.7513 ±0.0093 best at pub: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) paper 0.7513 ±0.0093 Δ at pub 0.0590
77	GraphTARIF Hyb baseline	0.8093 ±0.0057	This paper
88	GCN+TS GNN baseline	0.8031 ±0.0044	This paper
90	Polynormer GT baseline actual #90 vs paper #90	0.8026 ±0.0092	actual #90 vs paper #90 paper 0.8026 ±0.0092 prev best: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) Improvement within reported uncertainty; not highlighted. paper 0.8026 ±0.0092 Δ at pub 0.0073
117	Exphormer GT baseline actual #117 vs paper #117	0.7938 ±0.0062	actual #117 vs paper #117 paper 0.7938 ±0.0062 prev best: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) Improvement within reported uncertainty; not highlighted. paper 0.7938 ±0.0062 Δ at pub 0.0084
120	GraphGPS Hyb baseline actual #120 vs paper #150 30 places worse	0.7926 ±0.0057 best reported	actual #120 vs paper #150 30 places worse paper 0.7866 ±0.0049 best at pub: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.7866 ±0.0049 Δ at pub 0.0060
176	GOAT GT baseline actual #176 vs paper #176	0.7796 ±0.0063	actual #176 vs paper #176 paper 0.7796 ±0.0063 prev best: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) paper 0.7796 ±0.0063 Δ at pub 0.0096
237	MLP Trad baseline actual #237 vs paper #262 25 places worse	0.7469 ±0.0085 best reported	actual #237 vs paper #262 25 places worse paper 0.7096 ±0.0100 best at pub: View Space: Learning Representation across Arbitrary Graphs (ICML 2025) paper 0.7096 ±0.0100 Δ at pub 0.0373
249	MLP+TS Trad baseline	0.7248 ±0.0083	This paper

Heterophilic Graphs

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 GNN baseline actual #3 vs paper #229 226 places worse	0.8560 best reported	actual #3 vs paper #229 226 places worse paper 0.4387 ±0.0520 best at pub: The Heterophilic Snowflake Hypothesis: Training and Empowering GNNs for Heterophilic Graphs (KDD 2024) paper 0.4387 ±0.0520 Δ at pub 0.4173
25	FSGNN GNN baseline actual #25 vs paper #251 226 places worse	0.7827 ±0.0128 best reported	actual #25 vs paper #251 226 places worse paper 0.4061 ±0.0297 best at pub: Efficient Learning on Large Graphs using a Densifying Regularity Lemma (2025) best today: Directed Homophily-Aware Graph Neural Network (2025) paper 0.4061 ±0.0297 Δ at pub 0.3766 Δ today 0.3766
45	GloGNN GNN baseline actual #45 vs paper #282 237 places worse	0.7133 ±0.0242 best reported	actual #45 vs paper #282 237 places worse paper 0.2590 ±0.0358 best at pub: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) paper 0.2590 ±0.0358 Δ at pub 0.4543
52	GCN GNN baseline actual #52 vs paper #224 172 places worse	0.7033 best reported	actual #52 vs paper #224 172 places worse paper 0.4530 ±0.0230 best at pub: Understanding Heterophily for Graph Neural Networks (ICML 2024) paper 0.4530 ±0.0230 Δ at pub 0.2503
72	GPRGNN GNN baseline actual #72 vs paper #254 182 places worse	0.6748 ±0.0040 best reported	actual #72 vs paper #254 182 places worse paper 0.3993 ±0.0330 best at pub: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) paper 0.3993 ±0.0330 Δ at pub 0.2755
108	SAGE GNN baseline actual #108 vs paper #229 121 places worse	0.6340 ±0.0300 best reported	actual #108 vs paper #229 121 places worse paper 0.4432 ±0.0455 best at pub: Scale-aware Message Passing For Graph Node Classification (2024) paper 0.4432 ±0.0455 Δ at pub 0.1908
115	GraphSAGE code ↗ GNN global	0.6215 ±0.0042	Hierarchical Graph Transformer with Adaptive Node Sampling code ↗
133	H2GCN GNN baseline actual #133 vs paper #279 146 places worse	0.6030 ±0.0231 best reported	actual #133 vs paper #279 146 places worse paper 0.2675 ±0.0364 best at pub: Optimization-Induced Graph Implicit Nonlinear Diffusion (ICML 2022) paper 0.2675 ±0.0364 Δ at pub 0.3355
138	Polynormer GT baseline actual #138 vs paper #245 107 places worse	0.6000 ±0.0180 best reported	actual #138 vs paper #245 107 places worse paper 0.4197 ±0.0318 best at pub: Scale-aware Message Passing For Graph Node Classification (2024) paper 0.4197 ±0.0318 Δ at pub 0.1803
159	GraphGPS Hyb baseline actual #159 vs paper #246 87 places worse	0.5620 ±0.0380 best reported	actual #159 vs paper #246 87 places worse paper 0.4155 ±0.0391 best at pub: Scale-aware Message Passing For Graph Node Classification (2024) paper 0.4155 ±0.0391 Δ at pub 0.1465
169	MLP Trad baseline actual #169 vs paper #229 60 places worse	0.5429 best reported	actual #169 vs paper #229 60 places worse paper 0.4386 ±0.0523 best at pub: Understanding Heterophily for Graph Neural Networks (ICML 2024) paper 0.4386 ±0.0523 Δ at pub 0.1043
188	SGFormer GT baseline actual #188 vs paper #225 37 places worse	0.5140 ±0.0170 best reported	actual #188 vs paper #225 37 places worse paper 0.4521 ±0.0372 best at pub: Scale-aware Message Passing For Graph Node Classification (2024) paper 0.4521 ±0.0372 Δ at pub 0.0619
212	GCN+ReP GNN baseline	0.4757 ±0.0390	This paper
213	CoED GNN baseline	0.4727 ±0.0362	This paper
215	M3Dphormer GT baseline	0.4709 ±0.0405	This paper
225	GCN+TS GNN baseline	0.4527 ±0.0474	This paper
228	GAT+T Hyb baseline	0.4452 ±0.0369	This paper
230	MLP+TS Trad baseline	0.4386 ±0.0523	This paper
237	SAGE+T GNN baseline	0.4332 ±0.0467	This paper
267	NodeFormer GT baseline actual #267 vs paper #267	0.3638 ±0.0385	actual #267 vs paper #267 paper 0.3638 ±0.0385 prev best: Graph Homophily Booster: Rethinking the Role of Discrete Features on Heterophilic Graphs (2025) Improvement within reported uncertainty; not highlighted. paper 0.3638 ±0.0385 Δ at pub 0.0146
270	CPGNN GNN baseline	0.3300 ±0.0315	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
11	FSGNN GNN baseline actual #11 vs paper #205 194 places worse	0.7416 best reported	actual #11 vs paper #205 194 places worse paper 0.3592 ±0.0132 best at pub: Simplifying approach to Node Classification in Graph Neural Networks (2021) paper 0.3592 ±0.0132 Δ at pub 0.3824
19	GloGNN GNN baseline actual #19 vs paper #217 198 places worse	0.7121 ±0.0184 best reported	actual #19 vs paper #217 198 places worse paper 0.3511 ±0.0124 best at pub: Directed Homophily-Aware Graph Neural Network (2025) paper 0.3511 ±0.0124 Δ at pub 0.3610
39	H2GCN GNN baseline actual #39 vs paper #217 178 places worse	0.6190 ±0.0140 best reported	actual #39 vs paper #217 178 places worse paper 0.3510 ±0.0115 best at pub: Efficient Learning on Large Graphs using a Densifying Regularity Lemma (2025) paper 0.3510 ±0.0115 Δ at pub 0.2680
46	GCN GNN baseline actual #46 vs paper #124 78 places worse	0.6067 best reported	actual #46 vs paper #124 78 places worse paper 0.4425 ±0.0204 best at pub: Understanding Heterophily for Graph Neural Networks (ICML 2024) paper 0.4425 ±0.0204 Δ at pub 0.1642
61	GraphGPS Hyb baseline actual #61 vs paper #166 105 places worse	0.5690 ±0.0300 best reported	actual #61 vs paper #166 105 places worse paper 0.3981 ±0.0228 best at pub: Scale-aware Message Passing For Graph Node Classification (2024) paper 0.3981 ±0.0228 Δ at pub 0.1709
67	GAT GNN baseline actual #67 vs paper #150 83 places worse	0.5630 ±0.0220 best reported	actual #67 vs paper #150 83 places worse paper 0.4151 ±0.0234 best at pub: Restructuring Graphs for Higher Homophily via Adaptive Spectral Clustering (AAAI 2022) paper 0.4151 ±0.0234 Δ at pub 0.1479
76	GPRGNN GNN baseline actual #76 vs paper #173 97 places worse	0.5435 ±0.0087 best reported	actual #76 vs paper #173 97 places worse paper 0.3895 ±0.0199 best at pub: Directed Homophily-Aware Graph Neural Network (2025) paper 0.3895 ±0.0199 Δ at pub 0.1540
110	GCN+ReP GNN baseline	0.4589 ±0.0145	This paper
113	GraphTARIF Hyb baseline	0.4558 ±0.0191	This paper
114	CoED GNN baseline	0.4550 ±0.0162	This paper
120	SAGE GNN baseline actual #120 vs paper #160 40 places worse	0.4460 ±0.0130 best reported	actual #120 vs paper #160 40 places worse paper 0.4048 ±0.0290 best at pub: Scale-aware Message Passing For Graph Node Classification (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.4048 ±0.0290 Δ at pub 0.0412
121	GCN+TS GNN baseline	0.4457 ±0.0204	This paper
123	M3Dphormer GT baseline	0.4434 ±0.0194	This paper
128	GraphSAGE GNN global	0.4378 ±0.0190	UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node …
138	SGFormer GT baseline actual #138 vs paper #138	0.4265 ±0.0241	actual #138 vs paper #138 paper 0.4265 ±0.0241 prev best: SCNode: Spatial and Contextual Coordinates for Graph Representation Learning (TMLR 2024) paper 0.4265 ±0.0241 Δ at pub 0.0684
146	Polynormer GT baseline actual #146 vs paper #146	0.4197 ±0.0214	actual #146 vs paper #146 paper 0.4197 ±0.0214 prev best: Scale-aware Message Passing For Graph Node Classification (2024) paper 0.4197 ±0.0214 Δ at pub 0.0347
151	SAGE+T GNN baseline	0.4132 ±0.0235	This paper
162	GAT+T Hyb baseline	0.4036 ±0.0166	This paper
171	MLP+TS Trad baseline	0.3939 ±0.0091	This paper
172	MLP Trad baseline actual #172 vs paper #172	0.3930 ±0.0079	actual #172 vs paper #172 paper 0.3930 ±0.0079 prev best: Understanding Heterophily for Graph Neural Networks (ICML 2024) paper 0.3930 ±0.0079 Δ at pub 0.0267
175	NodeFormer GT baseline actual #175 vs paper #175	0.3889 ±0.0267	actual #175 vs paper #175 paper 0.3889 ±0.0267 prev best: SCNode: Spatial and Contextual Coordinates for Graph Representation Learning (TMLR 2024) Improvement within reported uncertainty; not highlighted. paper 0.3889 ±0.0267 Δ at pub 0.0037
241	CPGNN GNN baseline	0.3004 ±0.0203	This paper

Heterophily Benchmark

Roman-empire

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	Dir-Poly GNN baseline	0.9451 ±0.0022	This paper
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 ↗
5	GraphTARIF Hyb baseline	0.9323 ±0.0038	This paper
7	Polynormer GT baseline actual #7 vs paper #7	0.9266 ±0.0060	actual #7 vs paper #7 paper 0.9266 ±0.0060 prev best: Directed Semi-Simplicial Learning with Applications to Brain Activity Decoding (2025) Improvement within reported uncertainty; not highlighted. paper 0.9266 ±0.0060 Δ at pub 0.0011
11	CoED GNN baseline	0.9217 ±0.0029	This paper
15	BuNN Hyb baseline	0.9175 ±0.0039	This paper
16	GCN+TS GNN baseline	0.9166 ±0.0020	This paper
23	SAGE+T GNN baseline	0.9127 ±0.0036	This paper
26	GraphSAGE code ↗ GNN global	0.9106 ±0.0027	Directed Semi-Simplicial Learning with Applications to Brain Activity Decoding code ↗
27	GAT+T Hyb baseline	0.9093 ±0.0020	This paper
61	GCN+ReP GNN baseline	0.8643 ±0.0074	This paper
64	SAGE GNN baseline actual #64 vs paper #179 115 places worse	0.8574 ±0.0067 best reported	actual #64 vs paper #179 115 places worse paper 0.5569 ±0.0023 best at pub: Graph Neural Networks for Graphs with Heterophily: A Survey (2022) best today: TANGO: Graph Neural Dynamics via Learned Energy and Tangential Flows (2025) paper 0.5569 ±0.0023 Δ at pub 0.3005 Δ today 0.3005
77	GCN GNN baseline actual #77 vs paper #180 103 places worse	0.8272 ±0.0082 best reported	actual #77 vs paper #180 103 places worse paper 0.5411 ±0.0034 best at pub: HyperAggregation: Aggregating over Graph Edges with Hypernetworks (2024) paper 0.5411 ±0.0034 Δ at pub 0.2861
78	GraphGPS Hyb baseline	0.8272 ±0.0068	This paper
83	GAT GNN baseline actual #83 vs paper #179 96 places worse	0.8145 ±0.0009 best reported	actual #83 vs paper #179 96 places worse paper 0.5559 ±0.0027 best at pub: Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax (2025) paper 0.5559 ±0.0027 Δ at pub 0.2586
90	SGFormer GT baseline actual #90 vs paper #90	0.8001 ±0.0044	actual #90 vs paper #90 paper 0.8001 ±0.0044 prev best: NTFormer: A Composite Node Tokenized Graph Transformer for Node Classification (2024) paper 0.8001 ±0.0044 Δ at pub 0.1462
92	FSGNN GNN baseline	0.7992 ±0.0056	This paper
109	NodeFormer GT baseline actual #109 vs paper #109	0.7483 ±0.0081	actual #109 vs paper #109 paper 0.7483 ±0.0081 prev best: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) paper 0.7483 ±0.0081 Δ at pub 0.1034
129	H2GCN GNN baseline actual #129 vs paper #164 35 places worse	0.6809 ±0.0029 best reported	actual #129 vs paper #164 35 places worse paper 0.6011 ±0.0052 best at pub: Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND (ICLR 2024) best today: Distributed-Order Fractional Graph Operating Network (NeurIPS 2024) paper 0.6011 ±0.0052 Δ at pub 0.0798 Δ today 0.0798
133	GPRGNN GNN baseline actual #133 vs paper #144 11 places worse	0.6704 ±0.0092 best reported	actual #133 vs paper #144 11 places worse paper 0.6485 ±0.0027 best at pub: NTFormer: A Composite Node Tokenized Graph Transformer for Node Classification (2024) paper 0.6485 ±0.0027 Δ at pub 0.0219
135	MLP Trad baseline actual #135 vs paper #138 3 places worse	0.6665 ±0.0047 best reported	actual #135 vs paper #138 3 places worse paper 0.6610 ±0.0044 best at pub: HyperAggregation: Aggregating over Graph Edges with Hypernetworks (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.6610 ±0.0044 Δ at pub 0.0055
138	MLP+TS Trad baseline	0.6612 ±0.0029	This paper
152	CPGNN GNN baseline	0.6396 ±0.0062	This paper
154	GloGNN GNN baseline actual #154 vs paper #167 13 places worse	0.6385 ±0.0049 best reported	actual #154 vs paper #167 13 places worse paper 0.5963 ±0.0069 best at pub: Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND (ICLR 2024) best today: Distributed-Order Fractional Graph Operating Network (NeurIPS 2024) paper 0.5963 ±0.0069 Δ at pub 0.0422 Δ today 0.0422

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 …
7	H2GCN GNN baseline actual #7 vs paper #196 189 places worse	0.5681 ±0.0090 best reported	actual #7 vs paper #196 189 places worse paper 0.3647 ±0.0023 best at pub: UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node Classification Tasks (2024) paper 0.3647 ±0.0023 Δ at pub 0.2034
8	SAGE+T GNN baseline	0.5672 ±0.0035	This paper
9	GraphTARIF Hyb baseline	0.5586 ±0.0042	This paper
11	GAT+T Hyb baseline	0.5573 ±0.0031	This paper
12	SAGE GNN baseline actual #12 vs paper #12	0.5569 ±0.0023	actual #12 vs paper #12 paper 0.5569 ±0.0023 prev best: Graph Neural Networks for Graphs with Heterophily: A Survey (2022) paper 0.5569 ±0.0023 Δ at pub 0.0206
14	GAT GNN baseline actual #14 vs paper #14	0.5559 ±0.0027	actual #14 vs paper #14 paper 0.5559 ±0.0027 prev best: Relieving the Over-Aggregating Effect in Graph Transformers (NeurIPS 2025) paper 0.5559 ±0.0027 Δ at pub 0.0289
15	GraphSAGE GNN global	0.5540 ±0.0020	UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node …
18	Polynormer GT baseline actual #18 vs paper #18	0.5496 ±0.0022	actual #18 vs paper #18 paper 0.5496 ±0.0022 prev best: Relieving the Over-Aggregating Effect in Graph Transformers (NeurIPS 2025) Improvement within reported uncertainty; not highlighted. paper 0.5496 ±0.0022 Δ at pub 0.0025
23	SGFormer GT baseline actual #23 vs paper #23	0.5414 ±0.0062	actual #23 vs paper #23 paper 0.5414 ±0.0062 prev best: Relieving the Over-Aggregating Effect in Graph Transformers (NeurIPS 2025) paper 0.5414 ±0.0062 Δ at pub 0.0176
24	GCN GNN baseline actual #24 vs paper #24	0.5411 ±0.0034	actual #24 vs paper #24 paper 0.5411 ±0.0034 prev best: UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node Classification Tasks (2024) Improvement within reported uncertainty; not highlighted. paper 0.5411 ±0.0034 Δ at pub 0.0031
26	GCN+TS GNN baseline	0.5406 ±0.0058	This paper
28	BuNN Hyb baseline	0.5374 ±0.0051	This paper
40	GraphGPS Hyb baseline actual #40 vs paper #40	0.5327 ±0.0066	actual #40 vs paper #40 paper 0.5327 ±0.0066 prev best: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) Improvement within reported uncertainty; not highlighted. paper 0.5327 ±0.0066 Δ at pub 0.0017
49	GCN+ReP GNN baseline	0.5275 ±0.0062	This paper
50	FSGNN GNN baseline	0.5274 ±0.0083	This paper
81	Dir-Poly GNN baseline	0.5073 ±0.0056	This paper
84	MLP Trad baseline actual #84 vs paper #108 24 places worse	0.5058 best reported	actual #84 vs paper #108 24 places worse paper 0.4885 ±0.0055 best at pub: Understanding Heterophily for Graph Neural Networks (ICML 2024) paper 0.4885 ±0.0055 Δ at pub 0.0173
101	MLP+TS Trad baseline	0.4950 ±0.0027	This paper
149	GPRGNN GNN baseline	0.4488 ±0.0034	This paper
154	CPGNN GNN baseline actual #154 vs paper #188 34 places worse	0.4436 ±0.0035 best reported	actual #154 vs paper #188 34 places worse paper 0.3979 ±0.0077 best at pub: Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND (ICLR 2024) best today: Distributed-Order Fractional Graph Operating Network (NeurIPS 2024) paper 0.3979 ±0.0077 Δ at pub 0.0457 Δ today 0.0457
160	NodeFormer GT baseline actual #160 vs paper #160	0.4386 ±0.0035 best reported	actual #160 vs paper #160 paper 0.4379 ±0.0057 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) Difference within reported uncertainty; not counted as underperformance. paper 0.4379 ±0.0057 Δ at pub 0.0007 Δ today 0.0007
196	GloGNN GNN baseline actual #196 vs paper #196	0.3728 ±0.0066 best reported	actual #196 vs paper #196 paper 0.3689 ±0.0014 best at pub: Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND (ICLR 2024) best today: Distributed-Order Fractional Graph Operating Network (NeurIPS 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.3689 ±0.0014 Δ at pub 0.0039 Δ today 0.0039

Minesweeper

Global leaderboard →

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

Sorted by Accuracy. Highlighted rows are from other papers.

#	Model	Accuracy ↑	Paper
1	GraphTARIF Hyb baseline	0.9903 ±0.0019	This paper
2	BuNN Hyb baseline	0.9899 ±0.0016	This paper
3	M3Dphormer GT baseline	0.9827 ±0.0020	This paper
5	GAT GNN baseline actual #5 vs paper #5	0.9786 ±0.0037	actual #5 vs paper #5 paper 0.9786 ±0.0037 prev best: Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing (2025) paper 0.9786 ±0.0037 Δ at pub 0.0585
6	GAT+T Hyb baseline	0.9786 ±0.0037	This paper
7	GCN+TS GNN baseline	0.9780 ±0.0020	This paper
8	Polynormer GT baseline actual #8 vs paper #8	0.9749 ±0.0048	actual #8 vs paper #8 paper 0.9749 ±0.0048 prev best: Neighbourhood Transformer: Switchable Attention for Monophily-Aware Graph Learning (2026) Improvement within reported uncertainty; not highlighted. paper 0.9749 ±0.0048 Δ at pub 0.0003
9	GCN GNN baseline actual #9 vs paper #9	0.9744 ±0.0017	actual #9 vs paper #9 paper 0.9744 ±0.0017 prev best: Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing (2025) paper 0.9744 ±0.0017 Δ at pub 0.0769
10	SAGE+T GNN baseline	0.9733 ±0.0094	This paper
11	SAGE GNN baseline	0.9709 ±0.0100	This paper
12	GCN+ReP GNN baseline	0.9605 ±0.0019	This paper
16	Dir-Poly GNN baseline	0.9374 ±0.0070	This paper
17	GraphGPS Hyb baseline actual #17 vs paper #22 5 places worse	0.9229 ±0.0061 best reported	actual #17 vs paper #22 5 places worse paper 0.9075 ±0.0089 best at pub: Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing (2025) paper 0.9075 ±0.0089 Δ at pub 0.0154
20	SGFormer GT baseline	0.9142 ±0.0041	This paper
24	FSGNN GNN baseline	0.9008 ±0.0070	This paper
25	H2GCN GNN baseline	0.8971 ±0.0031	This paper
27	NodeFormer GT baseline actual #27 vs paper #27	0.8771 ±0.0069	actual #27 vs paper #27 paper 0.8771 ±0.0069 prev best: Neighbourhood Transformer: Switchable Attention for Monophily-Aware Graph Learning (2026) Improvement within reported uncertainty; not highlighted. paper 0.8771 ±0.0069 Δ at pub 0.0100
31	GPRGNN GNN baseline	0.8624 ±0.0061	This paper
43	MLP Trad baseline actual #43 vs paper #58 15 places worse	0.8000 best reported	actual #43 vs paper #58 15 places worse paper 0.5106 ±0.0176 best at pub: HyperAggregation: Aggregating over Graph Edges with Hypernetworks (2024) best today: Learning Posterior Predictive Distributions for Node Classification from Synthetic Graph Priors (ICLR 2026) paper 0.5106 ±0.0176 Δ at pub 0.2894 Δ today 0.2894
57	CPGNN GNN baseline	0.5203 ±0.0546	This paper
58	GloGNN GNN baseline	0.5108 ±0.0123	This paper
59	MLP+TS Trad baseline	0.5097 ±0.0156	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