diff --git a/entries/ogbn_arxiv/__init__.py b/entries/ogbn_arxiv/__init__.py
new file mode 100644
index 0000000..f650687
--- /dev/null
+++ b/entries/ogbn_arxiv/__init__.py
@@ -0,0 +1,138 @@
+#! /usr/bin/env python
+# -*- coding: utf-8 -*-
+#
+# Copyright (C) 2026 Tiago de Paula Peixoto <tiago@skewed.de>
+#
+# This program is free software: you can redistribute it and/or modify it under
+# the terms of the GNU Affero General Public License as published by the Free
+# Software Foundation, either version 3 of the License, or (at your option) any
+# later version.
+#
+# This program is distributed in the hope that it will be useful, but WITHOUT
+# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+# FOR A PARTICULAR PURPOSE.  See the GNU Affero General Public License for more
+# details.
+#
+# You should have received a copy of the GNU Affero General Public License along
+# with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+
+import pyarrow
+import pyarrow.csv
+
+from .. import *
+
+
+title = "arXiv computer science citations (ogbn-arxiv, 1971-2020)"
+description = """A directed citation network of the computer science papers of the arXiv preprint server, as indexed by the Microsoft Academic Graph (MAG), from the `ogbn-arxiv` dataset of the Open Graph Benchmark (OGB), a benchmark for node classification (version 1, of May 2020). Each paper is labeled with its primary arXiv subject area (one of the 40 computer science categories, chosen by the authors and the arXiv moderators), and the publication years range from 1971 to 2020.
+
+An edge from paper A to paper B means that A cites B. There are no self-loops, and 8,444 pairs of papers cite each other.
+
+Each paper also has the node features of the benchmark: a 128-dimensional vector obtained by averaging the word2vec embeddings of the words in its title and abstract (the embeddings were trained by the OGB authors on the MAG corpus). The titles of the papers come from a separate file of the OGB authors, which also has their abstracts (left out). The titles are as normalized by MAG (in lower case, without punctuation).
+
+The network has the following properties:
+
+- `mag_id`: the identifier of the paper in the Microsoft Academic Graph.
+- `title`: the title of the paper, normalized.
+- `year`: the publication year of the paper.
+- `subject`: the primary arXiv subject area of the paper, e.g. `cs.LG` or `cs.AI`.
+- `features`: the 128-dimensional feature vector of the paper (the average of the word2vec embeddings of the words of its title and abstract)."""
+tags = ['Informational', 'Citation', 'Unweighted', 'Metadata']
+url = ['https://ogb.stanford.edu/docs/nodeprop/#ogbn-arxiv',
+       'https://snap.stanford.edu/ogb/data/']
+citation = [('W. Hu, M. Fey, M. Zitnik, Y. Dong, H. Ren, B. Liu, M. Catasta, J. Leskovec, "Open Graph Benchmark: Datasets for Machine Learning on Graphs." Advances in Neural Information Processing Systems 33, 22118-22133 (2020)', 'https://doi.org/10.48550/arXiv.2005.00687'),
+            ('K. Wang, Z. Shen, C. Huang, C.-H. Wu, Y. Dong, A. Kanakia, "Microsoft Academic Graph: When experts are not enough." Quantitative Science Studies 1(1), 396-413 (2020)', 'https://doi.org/10.1162/qss_a_00021')]
+icon_hash = None
+upstream_license = ('ODC-BY 1.0', 'https://opendatacommons.org/licenses/by/1-0/')
+upstream_prefix = 'https://snap.stanford.edu/ogb/data'
+files = [(('nodeproppred/arxiv.zip:arxiv/raw/edge.csv.gz',
+           'nodeproppred/arxiv.zip:arxiv/raw/node_year.csv.gz',
+           'nodeproppred/arxiv.zip:arxiv/raw/node-label.csv.gz',
+           'nodeproppred/arxiv.zip:arxiv/mapping/labelidx2arxivcategeory.csv.gz',
+           'nodeproppred/arxiv.zip:arxiv/mapping/nodeidx2paperid.csv.gz',
+           'misc/ogbn_arxiv/titleabs.tsv.gz:tar:titleabs.tsv',
+           'nodeproppred/arxiv.zip:arxiv/raw/node-feat.csv.gz'),
+          None, None)]
+
+def fetch_upstream(force=False):
+    return fetch_upstream_files(__name__.split(".")[-1], upstream_prefix, files,
+                                force)
+
+def read_csv_gz(name, fname):
+    "The lines of the gzipped CSV member `fname`."
+    with open_upstream_file(name, fname, "rb") as f:
+        with gzip.open(f, "rt") as gf:
+            return gf.read().splitlines()
+
+def set_features(g, name, fname, n):
+    """Store the node features of the gzipped CSV member `fname` (one line of
+    128 numbers per node, for the first `n` vertices) as the vector<double>
+    vertex property `features`, read in blocks."""
+    p = g.new_vp("vector<double>")
+    i = 0
+    with open_upstream_file(name, fname, "rb") as f, gzip.open(f) as gf:
+        reader = pyarrow.csv.open_csv(
+            gf,
+            read_options=pyarrow.csv.ReadOptions(autogenerate_column_names=True,
+                                                 block_size=1 << 26),
+            convert_options=pyarrow.csv.ConvertOptions(
+                column_types={f"f{j}": pyarrow.float64() for j in range(128)}))
+        for batch in reader:
+            assert batch.num_columns == 128 and i + batch.num_rows <= n
+            x = numpy.column_stack([c.to_numpy() for c in batch.columns])
+            for row in x:
+                p[i] = row
+                i += 1
+            del x
+    assert i == n, (i, n)
+    g.vp.features = p
+
+@cache_network()
+@coerce_props()
+@annotate()
+def parse(alts=None):
+    global files
+    name = __name__.split(".")[-1]
+    for fnames, alt, fmt in files:
+        if alts is not None and alt not in alts:
+            continue
+        edges = numpy.array([l.split(",") for l in read_csv_gz(name, fnames[0])],
+                            dtype=numpy.int64)
+        years = [int(l) for l in read_csv_gz(name, fnames[1])]
+        labels = [int(l) for l in read_csv_gz(name, fnames[2])]
+        # "label idx,arxiv category", e.g. "0,arxiv cs na"
+        cats = {}
+        for l in read_csv_gz(name, fnames[3])[1:]:
+            i, c = l.split(",")
+            c = c.split()
+            assert c[:2] == ["arxiv", "cs"], c
+            cats[int(i)] = f"cs.{c[2].upper()}"
+        # "node idx,paper id"
+        mag = {}
+        for l in read_csv_gz(name, fnames[4])[1:]:
+            i, p = l.split(",")
+            mag[int(i)] = int(p)
+        N = len(years)
+        assert len(labels) == N and sorted(mag) == list(range(N))
+
+        # "<MAG id>\t<title>\t<abstract>", for these and other papers
+        mids = set(mag.values())
+        titles = {}
+        with open_upstream_file(name, fnames[5], "r") as f:
+            for line in f:
+                row = line.rstrip("\n").split("\t")
+                if len(row) < 2 or not row[0].isdigit():
+                    continue
+                if int(row[0]) in mids:
+                    titles[int(row[0])] = row[1]
+
+        g = Graph(directed=True)
+        g.add_vertex(N)
+        g.vp.mag_id = g.new_vp("int64_t", vals=[mag[i] for i in range(N)])
+        g.vp.title = g.new_vp("string", vals=[titles.get(mag[i], "") for i in range(N)])
+        g.vp.year = g.new_vp("int", vals=years)
+        g.vp.subject = g.new_vp("string", vals=[cats[l] for l in labels])
+        # Edges: "<citing>,<cited>"
+        g.add_edge_list(edges)
+        set_features(g, name, fnames[6], N)
+        yield alt, g