diff --git a/.github/workflows/pytest.yml b/.github/workflows/pytest.yml
new file mode 100644
index 0000000..2c86851
--- /dev/null
+++ b/.github/workflows/pytest.yml
@@ -0,0 +1,74 @@
+name: Test
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: ["*"]
+  workflow_dispatch: # allows you to trigger manually
+
+# When this workflow is queued, automatically cancel any previous running
+# or pending jobs from the same branch
+concurrency:
+  group: pytest-${{ github.ref }}
+  cancel-in-progress: true
+
+defaults:
+  run:
+    shell: bash -l {0}
+
+jobs:
+  test:
+    name: ${{ matrix.os }} Python ${{ matrix.python-version }} NumPy ${{ matrix.numpy-version}}
+    runs-on: ${{ matrix.os}}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest]
+        python-version: ["3.11", "3.14"]
+        numpy-version: ["2.3.0", latest]
+        include:
+          # Test oldest supported Python and NumPy versions
+          - os: ubuntu-latest
+            python-version: "3.8"
+            numpy-version: "1.18.0"
+          # Test vs. NumPy nightly wheels
+          - os: ubuntu-latest
+            python-version: "3.14"
+            numpy-version: "nightly"
+          # Test issues re. preinstalled SSL certificates on different OSes
+          - os: windows-latest
+            python-version: "3.14"
+            numpy-version: latest
+          - os: macos-latest
+            python-version: "3.14"
+            numpy-version: latest
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v6
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v4
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install pinned NumPy
+        if: matrix.numpy-version != 'latest' && matrix.numpy-version != 'nightly'
+        run: python -m pip install numpy==${{ matrix.numpy-version }}
+
+      - name: Install nightly NumPy wheels
+        if: matrix.numpy-version == 'nightly'
+        run: pip install --pre --extra-index-url https://pypi.anaconda.org/scientific-python-nightly-wheels/simple/ numpy
+
+      - name: Install package
+        run: pip install .
+
+      - name: Smoke test
+        run: python -c "import ml_datasets"
+
+      - name: Install test dependencies
+        run: pip install pytest
+      
+      - name: Run tests
+        run: pytest
diff --git a/ml_datasets/__init__.py b/ml_datasets/__init__.py
index b4d6d89..56fe965 100644
--- a/ml_datasets/__init__.py
+++ b/ml_datasets/__init__.py
@@ -8,3 +8,5 @@
 from .loaders.universal_dependencies import ud_ancora_pos_tags, ud_ewtb_pos_tags
 from .loaders.dbpedia import dbpedia
 from .loaders.cmu import cmu
+from .loaders.cifar import cifar
+from .loaders.wikiner import wikiner
diff --git a/ml_datasets/loaders/__init__.py b/ml_datasets/loaders/__init__.py
index 747d385..07add6b 100644
--- a/ml_datasets/loaders/__init__.py
+++ b/ml_datasets/loaders/__init__.py
@@ -1,3 +1,4 @@
+from .cifar import cifar
 from .cmu import cmu
 from .dbpedia import dbpedia
 from .imdb import imdb
diff --git a/ml_datasets/loaders/reuters.py b/ml_datasets/loaders/reuters.py
index 367abfc..20428ef 100644
--- a/ml_datasets/loaders/reuters.py
+++ b/ml_datasets/loaders/reuters.py
@@ -4,8 +4,8 @@
 from ..util import get_file
 from .._registry import register_loader
 
-
-URL = "https://s3.amazonaws.com/text-datasets/reuters_word_index.pkl"
+URL = "https://s3.amazonaws.com/text-datasets/reuters.pkl"
+WORD_INDEX_URL = "https://s3.amazonaws.com/text-datasets/reuters_word_index.pkl"
 
 
 @register_loader("reuters")
@@ -15,7 +15,7 @@ def reuters():
 
 
 def get_word_index(path="reuters_word_index.pkl"):
-    path = get_file(path, origin=URL)
+    path = get_file(path, origin=WORD_INDEX_URL)
     f = open(path, "rb")
     data = pickle.load(f, encoding="latin1")
     f.close()
@@ -60,7 +60,7 @@ def load_reuters(
     # https://raw.githubusercontent.com/fchollet/keras/master/keras/datasets/mnist.py
     # Copyright Francois Chollet, Google, others (2015)
     # Under MIT license
-    path = get_file(path, origin="https://s3.amazonaws.com/text-datasets/reuters.pkl")
+    path = get_file(path, origin=URL)
     f = open(path, "rb")
     X, labels = pickle.load(f)
     f.close()
diff --git a/ml_datasets/test/__init__.py b/ml_datasets/test/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/ml_datasets/test/test_datasets.py b/ml_datasets/test/test_datasets.py
new file mode 100644
index 0000000..a27dbc0
--- /dev/null
+++ b/ml_datasets/test/test_datasets.py
@@ -0,0 +1,81 @@
+# TODO the tests below only verify that the various functions don't crash.
+# Expand them to test the actual output contents.
+
+import platform
+
+import pytest
+import numpy as np
+
+import ml_datasets
+
+NP_VERSION = tuple(int(x) for x in np.__version__.split(".")[:2])
+
+# FIXME warning on NumPy 2.4 when downloading pre-computed pickles:
+# Python or NumPy boolean but got `align=0`.
+# Did you mean to pass a tuple to create a subarray type? (Deprecated NumPy 2.4)
+if NP_VERSION >= (2, 4):
+    np_24_deprecation = pytest.mark.filterwarnings(
+        "ignore::numpy.exceptions.VisibleDeprecationWarning", 
+    
+    )
+else:
+    # Note: can't use `condition=NP_VERSION >= (2, 4)` on the decorator directly
+    # as numpy.exceptions did not exist in old NumPy versions.
+    np_24_deprecation = lambda x: x
+
+
+@np_24_deprecation
+def test_cifar():
+    (X_train, y_train), (X_test, y_test) = ml_datasets.cifar()
+
+
+@pytest.mark.skip(reason="very slow download")
+def test_cmu():
+    train, dev = ml_datasets.cmu()
+
+
+def test_dbpedia():
+    train, dev = ml_datasets.dbpedia()
+
+
+def test_imdb():
+    train, dev = ml_datasets.imdb()
+
+
+@np_24_deprecation
+def test_mnist():
+    (X_train, y_train), (X_test, y_test) = ml_datasets.mnist()
+
+
+@pytest.mark.xfail(reason="403 Forbidden")
+def test_quora_questions():
+    train, dev = ml_datasets.quora_questions()
+
+
+@np_24_deprecation
+def test_reuters():
+    (X_train, y_train), (X_test, y_test) = ml_datasets.reuters()
+
+
+@pytest.mark.xfail(platform.system() == "Windows", reason="path issues")
+def test_snli():
+    train, dev = ml_datasets.snli()
+
+
+@pytest.mark.xfail(reason="no default path")
+def test_stack_exchange():
+    train, dev = ml_datasets.stack_exchange()
+
+
+def test_ud_ancora_pos_tags():
+    (train_X, train_y), (dev_X, dev_y) = ml_datasets.ud_ancora_pos_tags()
+
+
+@pytest.mark.xfail(reason="str column where int expected")
+def test_ud_ewtb_pos_tags():
+    (train_X, train_y), (dev_X, dev_y) = ml_datasets.ud_ewtb_pos_tags()
+
+
+@pytest.mark.xfail(reason="no default path")
+def test_wikiner():
+    train, dev = ml_datasets.wikiner()
diff --git a/pyproject.toml b/pyproject.toml
new file mode 100644
index 0000000..12e2298
--- /dev/null
+++ b/pyproject.toml
@@ -0,0 +1,9 @@
+[tool.pytest.ini_options]
+addopts = "--strict-markers --strict-config -v -r sxfE --color=yes --durations=10"
+xfail_strict = true
+filterwarnings = [
+    "error",
+    # FIXME spurious random download warnings; will cause trouble in downstream CI
+    "ignore:Implicitly cleaning up <HTTPError 403:ResourceWarning",
+    "ignore:unclosed <socket.socket:ResourceWarning",
+]
diff --git a/requirements.txt b/requirements.txt
index 0b76345..ebd4257 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,5 +1,5 @@
 cloudpickle>=2.2
-numpy>=1.7.0
+numpy>=1.18
 scipy>=1.7.0
 tqdm>=4.10.0,<5.0.0
 # Our libraries
diff --git a/setup.cfg b/setup.cfg
index 9d40de5..30ea6a7 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -11,10 +11,10 @@ long_description_content_type = text/markdown
 [options]
 zip_safe = true
 include_package_data = true
-python_requires = >=3.6
+python_requires = >=3.8
 install_requires =
     cloudpickle>=2.2
-    numpy>=1.7.0
+    numpy>=1.18
     tqdm>=4.10.0,<5.0.0
     # Our libraries
     srsly>=1.0.1,<4.0.0