diff --git a/.github/workflows/python-compatibility.yml b/.github/workflows/python-compatibility.yml
new file mode 100644
index 0000000..efbe8d5
--- /dev/null
+++ b/.github/workflows/python-compatibility.yml
@@ -0,0 +1,31 @@
+name: Python compatibility
+
+on:
+  push:
+    branches: [master]
+  pull_request:
+
+permissions:
+  contents: read
+
+jobs:
+  cpu:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.13", "3.14"]
+    steps:
+      - uses: actions/checkout@v7
+      - uses: actions/setup-python@v7
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Build distributions
+        run: |
+          python -m pip install --upgrade pip build
+          python -m build
+      - name: Install wheel and test dependencies
+        run: python -m pip install dist/*.whl pytest
+      - name: Run test suite
+        working-directory: ${{ runner.temp }}
+        run: python -m pytest -q -rs "$GITHUB_WORKSPACE/tests"
diff --git a/README.md b/README.md
index 3e6e101..f78e51d 100644
--- a/README.md
+++ b/README.md
@@ -46,11 +46,16 @@ see [Tutorial_2_Advanced_multioutput](https://github.com/AILab-MLTools/Py-Boost/
 
 ## Installation
 
-Before installing py-boost via pip, you should have cupy installed. You can use:
+Py-Boost supports Python 3.8–3.14. Install the CuPy distribution compatible
+with your Python and CUDA versions before installing Py-Boost.
 
-`pip install -U cupy-cuda110 py-boost`
+On Python 3.14, use Numba 0.63 or later and CuPy 14 or later. For example,
+with CUDA 12.x:
 
-**Note**: replace with your cuda version! For details, see [this guide](https://docs.cupy.dev/en/stable/install.html).
+`pip install -U "cupy-cuda12x>=14" py-boost`
+
+For other Python or CUDA versions, choose the appropriate CuPy package from the
+[CuPy installation guide](https://docs.cupy.dev/en/stable/install.html).
 
 ## Quick tour
 
diff --git a/build_package.sh b/build_package.sh
index 782b52b..fbb4ee7 100644
--- a/build_package.sh
+++ b/build_package.sh
@@ -6,7 +6,7 @@ source ./py_boost_venv/bin/activate
 
 pip install -U pip
 pip install -U poetry
-pip install -U cupy-cuda110
+pip install -U "${CUPY_PACKAGE:-cupy-cuda12x}"
 
 poetry lock
 poetry install
diff --git a/pyproject.toml b/pyproject.toml
index b9e045f..f5116b0 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -19,6 +19,7 @@ classifiers = [
     "Programming Language :: Python :: 3.11",
     "Programming Language :: Python :: 3.12",
     "Programming Language :: Python :: 3.13",
+    "Programming Language :: Python :: 3.14",
     "Operating System :: OS Independent",
     "Intended Audience :: Science/Research",
     "Development Status :: 3 - Alpha",
@@ -29,7 +30,7 @@ classifiers = [
 
 [tool.poetry.dependencies]
 
-python = ">=3.8, <3.14"
+python = ">=3.8, <3.15"
 
 scikit-learn = ">=1"
 numpy = "*"
diff --git a/tests/test_quantization.py b/tests/test_quantization.py
new file mode 100644
index 0000000..e5dd286
--- /dev/null
+++ b/tests/test_quantization.py
@@ -0,0 +1,38 @@
+"""CPU coverage for the Numba kernels used by all training paths."""
+
+import numpy as np
+import pytest
+
+from py_boost.quantization.base import (
+    QuantileQuantizer,
+    UniformQuantizer,
+    UniquantQuantizer,
+)
+
+
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+@pytest.mark.parametrize("quantizer_cls", [
+    QuantileQuantizer, UniformQuantizer, UniquantQuantizer,
+])
+def test_quantization_matches_borders(quantizer_cls, dtype):
+    # More distinct values than bins exercises each quantizer's binning kernel.
+    X = np.linspace(-10, 10, 128, dtype=dtype).reshape(32, 4)
+    X[0] = np.nan
+    quantizer = quantizer_cls(max_bin=8)
+
+    encoded = quantizer.fit_transform(X)
+    borders = quantizer.get_borders()
+
+    assert encoded.dtype == np.uint8
+    assert encoded.shape == X.shape
+    assert quantizer.get_max_bin() <= 8
+    for column, column_borders in enumerate(borders):
+        assert np.isneginf(column_borders[0])
+        assert np.isposinf(column_borders[-1])
+        assert np.all(np.diff(column_borders) > 0)
+        valid = ~np.isnan(X[:, column])
+        expected = np.zeros(X.shape[0], dtype=np.uint8)
+        expected[valid] = np.searchsorted(column_borders, X[valid, column])
+        np.testing.assert_array_equal(encoded[:, column], expected)
+
+    np.testing.assert_array_equal(quantizer.transform(X), encoded)