diff --git a/.github/workflows/tests.yml b/.github/workflows/tests.yml
index 9b943f25c..c5430a54c 100644
--- a/.github/workflows/tests.yml
+++ b/.github/workflows/tests.yml
@@ -2,19 +2,26 @@ name: tests
 
 on:
   push:
+    branches: [main, v8.3.x]
     paths-ignore:
       - "website/**"
       - "*.md"
   pull_request:
-    types: [opened, synchronize, reopened, edited]
+    branches: ["*"]
     paths-ignore:
       - "website/**"
       - "*.md"
+  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: tests-${{ github.ref }}
+  cancel-in-progress: true
 
 jobs:
   validate:
     name: Validate
-    if: github.repository_owner == 'explosion'
     runs-on: ubuntu-latest
     steps:
       - name: Check out repo
@@ -35,26 +42,31 @@ jobs:
           python -m isort thinc --check
       - name: flake8
         run: |
-          python -m pip install flake8==5.0.4
+          python -m pip install flake8 -c requirements.txt
           python -m flake8 thinc --count --select=E901,E999,F821,F822,F823,W605 --show-source --statistics
+      - name: mypy
+        run: |
+          python -m pip install mypy -c requirements.txt
+          python -m pip install catalogue confection numpy packaging pydantic
+          python -m mypy thinc --no-implicit-reexport
+
   tests:
-    name: Test
-    needs: Validate
-    if: github.repository_owner == 'explosion'
+    name: ${{ matrix.os }} - Python ${{ matrix.python_version }}
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest, windows-latest, macos-15-intel]
-        python_version: ["3.12"]
-        include:
-          - os: windows-latest
-            python_version: "3.10"
-          - os: macos-15-intel
+        os:
+          - ubuntu-latest
+          - ubuntu-24.04-arm
+          - macos-15-intel
+          - macos-latest
+          - windows-latest
+          - windows-11-arm
+        python_version: ["3.10", "3.11", "3.12", "3.13"]
+        exclude:
+          - os: windows-11-arm
             python_version: "3.10"
-          - os: ubuntu-latest
-            python_version: "3.11"
-          - os: windows-latest
-            python_version: "3.11"
+
     runs-on: ${{ matrix.os }}
     env:
       NOTEBOOK_KERNEL: "thinc-notebook-tests"
@@ -68,26 +80,11 @@ jobs:
         with:
           python-version: ${{ matrix.python_version }}
 
-      - name: Install dependencies
-        run: |
-          python -m pip install --upgrade pip setuptools wheel
-          pip install -r requirements.txt
-
-      - name: Build sdist
-        run: |
-          # Remove the '.eggs' directory in case it's not empty
-          # due to setuptools quirks
-          rm -rf .eggs
-          python setup.py build_ext --inplace
-          rm -rf .eggs
-          python setup.py sdist --formats=gztar
-        shell: bash
+      - name: Install build dependencies
+        run: python -m pip install --upgrade build pip wheel
 
-      - name: Run mypy
-        run: python -m mypy thinc --no-implicit-reexport
-        if: |
-          matrix.python_version != '3.6' &&
-          matrix.python_version != '3.7'
+      - name: Build sdist and wheel
+        run: python -m build
 
       - name: Delete source directory
         run: rm -rf thinc
@@ -99,28 +96,24 @@ jobs:
           pip freeze --exclude pywin32 > installed.txt
           pip uninstall -y -r installed.txt
 
-      - name: Install from sdist
-        run: |
-          SDIST=$(python -c "import os;print(os.listdir('./dist')[-1])" 2>&1)
-          PIP_CONSTRAINT="build-constraints.txt" pip install dist/$SDIST
+      - name: Install from wheel
+        run: pip install dist/*.whl
         shell: bash
 
       - name: Test import
         run: python -c "import thinc"
 
       - name: Install test requirements
-        run: |
-          pip install -r requirements.txt
+        run: pip install -r requirements.txt
 
       - name: Install notebook test requirements
-        run: |
-          pip install ipykernel pydot graphviz
-          python -m ipykernel install --name thinc-notebook-tests --user
-        if: matrix.python_version != '3.12'
+        run: python -m ipykernel install --name thinc-notebook-tests --user
+
+      - name: List installed packages
+        run: python -m pip list
 
       - name: Run tests without extras
-        run: |
-          python -m pytest --pyargs thinc -Werror --cov=thinc --cov-report=term
+        run: python -m pytest --pyargs thinc --cov=thinc --cov-report=term
 
       # Notes on numpy requirements hacks:
       # 1. torch does not have a direct numpy requirement but is compiled
@@ -156,8 +149,8 @@ jobs:
           pip uninstall -y tensorflow
           pip install "thinc-apple-ops>=1.0.0,<2.0.0"
           python -m pytest --pyargs thinc_apple_ops
-        if: matrix.os == 'macos-15-intel' && matrix.python_version == '3.10'
+        if: runner.os == 'macOS' && matrix.python_version == '3.10'
 
       - name: Run tests with thinc-apple-ops
         run: python -m pytest --pyargs thinc
-        if: matrix.os == 'macos-15-intel' && matrix.python_version == '3.10'
+        if: runner.os == 'macOS' && matrix.python_version == '3.10'
diff --git a/README.md b/README.md
index 47dc81701..c53ce346b 100644
--- a/README.md
+++ b/README.md
@@ -36,8 +36,8 @@ their favorite framework.
 
 ## 🚀 Quickstart
 
-Thinc is compatible with **Python 3.6+** and runs on **Linux**, **macOS** and
-**Windows**. The latest releases with binary wheels are available from
+Thinc runs on **Linux**, **macOS** and **Windows**.
+The latest releases with binary wheels are available from
 [pip](https://pypi.python.org/pypi/thinc). Before you install Thinc and its
 dependencies, make sure that your `pip`, `setuptools` and `wheel` are up to
 date. For the most recent releases, pip 19.3 or newer is recommended.
@@ -121,8 +121,8 @@ the button next to the notebook name.
 
 Thinc uses [`black`](https://github.com/psf/black) for auto-formatting,
 [`flake8`](http://flake8.pycqa.org/en/latest/) for linting and
-[`mypy`](https://mypy.readthedocs.io/en/latest/) for type checking. All code is
-written compatible with **Python 3.6+**, with type hints wherever possible. See
+[`mypy`](https://mypy.readthedocs.io/en/latest/) for type checking. All code
+includes type hints wherever possible. See
 the [type reference](https://thinc.ai/docs/api-types) for more details on
 Thinc's custom types.
 
diff --git a/build-constraints.txt b/build-constraints.txt
deleted file mode 100644
index ccad7ef35..000000000
--- a/build-constraints.txt
+++ /dev/null
@@ -1,2 +0,0 @@
-# build version constraints for use with wheelwright + multibuild
-numpy>=2.0.0,<3.0.0
diff --git a/pyproject.toml b/pyproject.toml
index bcb8a68d7..243a60209 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -63,3 +63,19 @@ repair-wheel-command = "delocate-wheel --require-archs {delocate_archs} -w {dest
 
 [tool.isort]
 profile = "black"
+
+[tool.pytest.ini_options]
+addopts = "--strict-markers --strict-config -v -r sxfE --color=yes --durations=10"
+xfail_strict = true
+markers = [
+    "slow: test with long runtime",
+]
+filterwarnings = [
+    "error",
+    # This is triggered by pytest -p thinc.tests.enable_tensorflow
+    # Not an issue as the coverage upon import is recorded during the
+    # pytest run "without extras".
+    "ignore:Module thinc was previously imported, but not measured:coverage.exceptions.CoverageWarning",
+    # https://github.com/coveragepy/coveragepy/issues/1790
+    "ignore:Plugin file tracers:coverage.exceptions.CoverageWarning",
+]
diff --git a/requirements.txt b/requirements.txt
index 6e93ca62c..9fc46fdc4 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -3,36 +3,29 @@ murmurhash>=1.0.2,<1.1.0
 cymem>=2.0.2,<2.1.0
 preshed>=3.0.2,<3.1.0
 blis>=1.3.0,<1.4.0
-srsly>=2.4.0,<3.0.0
+srsly>=2.4.0,<3.1.0
 wasabi>=0.8.1,<1.2.0
 catalogue>=2.0.4,<2.1.0
-confection>=0.0.1,<1.0.0
-ml_datasets>=0.2.0,<0.3.0; python_version < "3.11"
+confection>=0.0.1,<1.1.0
+ml_datasets>=0.2.0,<0.3.0
 # Third-party dependencies
 pydantic>=2.0.0,<3.0.0
 numpy>=2.0.0,<3.0.0
 packaging>=20.0
-# Backports of modern Python features
-dataclasses>=0.6,<1.0; python_version < "3.7"
-typing_extensions>=3.7.4.1,<4.5.0; python_version < "3.8"
-contextvars>=2.4,<3; python_version < "3.7"
 # Development dependencies
-cython>=0.29.0
-hypothesis>=3.27.0,<6.72.2
+cython>=3.0,<4.0
+hypothesis>=3.27.0,<6.149
 pytest>=5.2.0,!=7.1.0
-pytest-cov>=2.7.0,<5.0.0
+pytest-cov>=2.7.0,<8.0.0
 coverage>=5.0.0,<8.0.0
-mock>=2.0.0,<3.0.0
-flake8>=3.5.0,<3.6.0
-mypy>=1.5.0,<1.6.0; platform_machine != "aarch64" and python_version >= "3.8"
-types-mock>=0.1.1
-types-contextvars>=0.1.2; python_version < "3.7"
-types-dataclasses>=0.1.3; python_version < "3.7"
-importlib_resources; python_version < "3.7"
+flake8==5.0.4
+mypy>=1.5.0,<1.6.0; platform_machine != "aarch64"
 # Executing notebook tests
-ipykernel>=5.1.4,<5.2.0
-nbconvert>=5.6.1,<6.5.0
-nbformat>=5.0.4,<5.2.0
+ipykernel>=5.1.4,<7.2
+pydot
+graphviz
+nbconvert>=5.6.1,<7.17
+nbformat>=5.0.4,<5.11
 # Test to_disk/from_disk against pathlib.Path subclasses
 pathy>=0.3.5
 black>=22.0,<23.0
diff --git a/setup.cfg b/setup.cfg
index 173efcf65..85057c38e 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -17,10 +17,6 @@ classifiers =
     Operating System :: Microsoft :: Windows
     Programming Language :: Cython
     Programming Language :: Python :: 3
-    Programming Language :: Python :: 3.6
-    Programming Language :: Python :: 3.7
-    Programming Language :: Python :: 3.8
-    Programming Language :: Python :: 3.9
     Programming Language :: Python :: 3.10
     Programming Language :: Python :: 3.11
     Programming Language :: Python :: 3.12
@@ -46,19 +42,15 @@ install_requires =
     cymem>=2.0.2,<2.1.0
     preshed>=3.0.2,<3.1.0
     wasabi>=0.8.1,<1.2.0
-    srsly>=2.4.0,<3.0.0
+    srsly>=2.4.0,<3.1.0
     catalogue>=2.0.4,<2.1.0
-    confection>=0.0.1,<1.0.0
+    confection>=0.0.1,<1.1.0
     # Third-party dependencies
     setuptools
-    numpy>=1.19.0,<3.0.0
+    numpy>=1.21.0,<3.0.0
     pydantic>=2.0.0,<3.0.0
     packaging>=20.0
-    # Backports of modern Python features
-    dataclasses>=0.6,<1.0; python_version < "3.7"
-    typing_extensions>=3.7.4.1,<5.0.0; python_version < "3.8"
-    contextvars>=2.4,<3; python_version < "3.7"
-    
+
 [options.entry_points]
 pytest_randomly.random_seeder =
     thinc = thinc.api:fix_random_seed
diff --git a/thinc/tests/layers/test_basic_tagger.py b/thinc/tests/layers/test_basic_tagger.py
index 3bc772940..3131ebb91 100644
--- a/thinc/tests/layers/test_basic_tagger.py
+++ b/thinc/tests/layers/test_basic_tagger.py
@@ -63,6 +63,7 @@ def get_shuffled_batches(Xs, Ys, batch_size):
 @pytest.mark.parametrize(
     ("depth", "width", "vector_width", "nb_epoch"), [(2, 32, 16, 5)]
 )
+@pytest.mark.xfail(reason="Flaky live download from the internet", strict=False)
 def test_small_end_to_end(depth, width, vector_width, nb_epoch, create_model, ancora):
     (train_X, train_Y), (dev_X, dev_Y) = ancora
     batch_size = 8
diff --git a/thinc/tests/layers/test_linear.py b/thinc/tests/layers/test_linear.py
index b1752fba2..e51e1dda1 100644
--- a/thinc/tests/layers/test_linear.py
+++ b/thinc/tests/layers/test_linear.py
@@ -1,7 +1,8 @@
+from unittest.mock import MagicMock
+
 import numpy
 import pytest
 from hypothesis import given, settings
-from mock import MagicMock
 from numpy.testing import assert_allclose
 
 from thinc.api import SGD, Dropout, Linear, chain
diff --git a/thinc/tests/layers/test_mnist.py b/thinc/tests/layers/test_mnist.py
index 060007cfd..8fdb74fa3 100644
--- a/thinc/tests/layers/test_mnist.py
+++ b/thinc/tests/layers/test_mnist.py
@@ -1,3 +1,5 @@
+import platform
+
 import pytest
 
 from thinc.api import (
@@ -79,6 +81,11 @@ def create_model(request):
 
 
 @pytest.mark.slow
+@pytest.mark.xfail(
+    platform.system() == "Darwin",
+    reason="SSL: CERTIFICATE_VERIFY_FAILED",
+    strict=False,  # Works on macos-15-intel Python 3.10, for some reason
+)
 @pytest.mark.parametrize(("width", "nb_epoch", "min_score"), [(32, 20, 0.8)])
 def test_small_end_to_end(width, nb_epoch, min_score, create_model, mnist):
     batch_size = 128
diff --git a/thinc/tests/layers/test_with_debug.py b/thinc/tests/layers/test_with_debug.py
index 3f65a3ac3..3e31c71f0 100644
--- a/thinc/tests/layers/test_with_debug.py
+++ b/thinc/tests/layers/test_with_debug.py
@@ -1,4 +1,4 @@
-from mock import MagicMock
+from unittest.mock import MagicMock
 
 from thinc.api import Linear, with_debug
 
diff --git a/thinc/tests/model/test_model.py b/thinc/tests/model/test_model.py
index f93b46c8c..2284e93b2 100644
--- a/thinc/tests/model/test_model.py
+++ b/thinc/tests/model/test_model.py
@@ -1,3 +1,4 @@
+import platform
 import threading
 import time
 from collections import Counter
@@ -513,6 +514,11 @@ def dummy_model(name, layers):
     assert a.layers[1].get_ref("y") == y_debug
 
 
+@pytest.mark.xfail(
+    platform.system() == "Darwin",
+    reason="SSL: CERTIFICATE_VERIFY_FAILED",
+    strict=False,  # Works on macos-15-intel Python 3.10, for some reason
+)
 def test_with_debug():
     pytest.importorskip("ml_datasets")
     import ml_datasets
diff --git a/thinc/tests/test_types.py b/thinc/tests/test_types.py
index d92446980..1a3310d30 100644
--- a/thinc/tests/test_types.py
+++ b/thinc/tests/test_types.py
@@ -19,7 +19,10 @@
 )
 
 
-@pytest.mark.xfail(reason="Validation currently disabled")
+@pytest.mark.xfail(
+    reason="Validation currently disabled",
+    strict=False,  # Some parameters pass, other fail
+)
 @pytest.mark.parametrize(
     "arr,arr_type",
     [