AI Paper Atlas

Research map / Benchmarks / PCQM4Mv2

PCQM4Mv2 leaderboard

10 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on PCQM4Mv2. Best MAE (val): 0.0734 (GeoMFormer, ICML 2024).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

PCQM4Mv2 · MAE (val)

10 papers, each with its best reported number for its own method; lower is better.

#MAE ↓MethodPaperTopic
10.0734GeoMFormerGeoMFormer: A General Architecture for Geometric Molecular Representation Learning
ICML 2024 · Table: Results on PCQM4Mv2
RNNs and normalization
20.0792GraphGPT-B48GraphGPT: Generative Pre-trained Graph Eulerian Transformer
ICML 2025 · Table: PCQM4Mv2 results
Graph self-supervised learning
30.084ETTowards Principled Graph Transformers
NeurIPS 2024 · PCQM4Mv2 table
GNN architectures
40.0858GPS-mediumRecipe for a General, Powerful, Scalable Graph Transformer
NeurIPS 2022 · PCQM4Mv2 results table
GNN architectures
50.0859GRITGraph Inductive Biases in Transformers without Message Passing
ICML 2023 · Table 4
GNN architectures
60.087(2,1) + LPEAligning Transformers with Weisfeiler-Leman
ICML 2024 · Table: Comparison of (2,1) to
GNN architectures
70.0889MPNN + VN + NoPE (medium)On the Connection Between MPNN and Graph Transformer
ICML 2023 · Table (PCQM4Mv2)
GNN architectures
80.0898GINE+RWSE
GINE
Position: Graph Learning Will Lose Relevance Due To Poor Benchmarks
ICML 2025 · Table: Evaluation results on t
Graph self-supervised learning
90.091TokenGT (Lap)Pure Transformers are Powerful Graph Learners
NeurIPS 2022 · Table: PCQM4Mv2 large-scale gr
GNN architectures
100.098SEM-IL (GIN)
GIN
Improving Compositional Generalization using Iterated Learning and Simplicial Embeddings
NeurIPS 2023 · Table: Downstream performance
Recognition and long-tail

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.