diff --git a/.github/workflows/benchmarks.yml b/.github/workflows/benchmarks.yml
new file mode 100644
index 000000000..e818a920d
--- /dev/null
+++ b/.github/workflows/benchmarks.yml
@@ -0,0 +1,132 @@
+# ------------------------------------------------------------------------------
+#  Copyright Matt Borland 2026.
+#  Distributed under the Boost Software License,
+#  Version 1.0. (See accompanying file LICENSE_1_0.txt
+#  or copy at http://www.boost.org/LICENSE_1_0.txt)
+# ------------------------------------------------------------------------------
+#
+# Runs the Boost.Decimal benchmarks (test/benchmarks.cpp) in release mode on a
+# spread of native runners so that performance numbers, and any regressions in
+# them, are captured on every run. The benchmark target is a run-fail test that
+# always returns non-zero, so b2 reports success and the timing output is read
+# back from the captured .output file.
+
+name: Run Benchmarks
+
+on:
+  push:
+    branches:
+      - master
+      - develop
+      - feature/**
+  pull_request:
+    types: [opened, synchronize, reopened]
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ format('{0}:{1}:benchmarks', github.repository, github.ref) }}
+  cancel-in-progress: true
+
+env:
+  GIT_FETCH_JOBS: 8
+
+jobs:
+  linux:
+    name: Linux GCC ${{ matrix.address_model }}-bit
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        address_model: [ 32, 64 ]
+    defaults:
+      run:
+        shell: bash
+    steps:
+      - uses: actions/checkout@v6
+      - name: Install packages
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y g++-14 g++-14-multilib
+      - name: Setup Boost
+        run: |
+          LIBRARY=${GITHUB_REPOSITORY#*/}
+          echo "LIBRARY=$LIBRARY" >> $GITHUB_ENV
+          REF=${GITHUB_BASE_REF:-$GITHUB_REF}
+          REF=${REF#refs/heads/}
+          BOOST_BRANCH=develop && [ "$REF" = "master" ] && BOOST_BRANCH=master || true
+          echo "BOOST_BRANCH: $BOOST_BRANCH"
+          cd ..
+          git clone -b "$BOOST_BRANCH" --depth 1 "https://github.com/boostorg/boost.git" boost-root
+          cd boost-root
+          mkdir -p libs/$LIBRARY
+          cp -r "$GITHUB_WORKSPACE"/* libs/$LIBRARY
+          git submodule update --init tools/boostdep
+          python tools/boostdep/depinst/depinst.py --git_args "--jobs $GIT_FETCH_JOBS" $LIBRARY
+          ./bootstrap.sh
+          ./b2 headers
+          echo "using gcc : : g++-14 ;" > ~/user-config.jam
+      - name: Run benchmarks (release)
+        run: |
+          cd ../boost-root
+          ./b2 -j$(nproc) toolset=gcc cxxstd=20 variant=release address-model=${{ matrix.address_model }} \
+            define=BOOST_DECIMAL_RUN_BENCHMARKS libs/$LIBRARY/test//benchmarks
+          echo "==================== Benchmark results (Linux GCC ${{ matrix.address_model }}-bit) ===================="
+          find bin.v2 -path '*release*' -name 'benchmarks.output' -exec cat {} +
+          find bin.v2 -path '*release*' -name 'benchmarks.output' -exec cp {} "$GITHUB_WORKSPACE/benchmarks-linux-gcc-${{ matrix.address_model }}-bit.txt" \;
+      - name: Upload benchmark output
+        if: always()
+        uses: actions/upload-artifact@v6
+        with:
+          name: benchmarks-linux-gcc-${{ matrix.address_model }}-bit
+          path: benchmarks-linux-gcc-${{ matrix.address_model }}-bit.txt
+          if-no-files-found: warn
+
+  windows:
+    name: ${{ matrix.name }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - name: Windows x64 MSVC
+            os: windows-latest
+            arch: x64
+          - name: Windows ARM64 MSVC
+            os: windows-11-arm
+            arch: arm64
+    defaults:
+      run:
+        shell: cmd
+    steps:
+      - uses: actions/checkout@v6
+      - name: Setup Boost
+        run: |
+          for /f %%i in ("%GITHUB_REPOSITORY%") do set LIBRARY=%%~nxi
+          echo LIBRARY=%LIBRARY%>>%GITHUB_ENV%
+          if "%GITHUB_BASE_REF%" == "" set GITHUB_BASE_REF=%GITHUB_REF%
+          set BOOST_BRANCH=develop
+          for /f %%i in ("%GITHUB_BASE_REF%") do if "%%~nxi" == "master" set BOOST_BRANCH=master
+          echo BOOST_BRANCH: %BOOST_BRANCH%
+          cd ..
+          git clone -b %BOOST_BRANCH% --depth 1 https://github.com/boostorg/boost.git boost-root
+          cd boost-root
+          xcopy /s /e /q %GITHUB_WORKSPACE% libs\%LIBRARY%\
+          git submodule update --init tools/boostdep
+          python tools/boostdep/depinst/depinst.py --git_args "--jobs 3" %LIBRARY%
+          cmd /c bootstrap
+          b2 -d0 headers
+      - name: Run benchmarks (release)
+        run: |
+          cd ../boost-root
+          b2 -j3 libs/%LIBRARY%/test//benchmarks toolset=msvc cxxstd=latest address-model=64 variant=release define=BOOST_DECIMAL_RUN_BENCHMARKS embed-manifest-via=linker || exit /b 1
+          echo ==================== Benchmark results (%RUNNER_OS% %RUNNER_ARCH%) ====================
+          for /f "delims=" %%f in ('dir /s /b bin.v2\benchmarks.output 2^>nul') do @type "%%f"
+          for /f "delims=" %%f in ('dir /s /b bin.v2\benchmarks.output 2^>nul') do @copy /y "%%f" "%GITHUB_WORKSPACE%\benchmarks-windows-${{ matrix.arch }}.txt" >nul
+          exit /b 0
+      - name: Upload benchmark output
+        if: always()
+        uses: actions/upload-artifact@v6
+        with:
+          name: benchmarks-windows-${{ matrix.arch }}
+          path: benchmarks-windows-${{ matrix.arch }}.txt
+          if-no-files-found: warn
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index c8c837340..b9fe14c11 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -1125,60 +1125,3 @@ jobs:
         run: |
           cd ../boost-root/__build__
           ctest --output-on-failure --no-tests=error
-
-  cuda-cmake-test:
-    strategy:
-      fail-fast: false
-
-    runs-on: ubuntu-24.04
-
-    steps:
-      - uses: Jimver/cuda-toolkit@v0.2.30
-        id: cuda-toolkit
-        with:
-          cuda: '12.8.0'
-          method: 'network'
-          sub-packages: '["nvcc"]'
-
-      - name: Output CUDA information
-        run: |
-          echo "Installed cuda version is: ${{steps.cuda-toolkit.outputs.cuda}}"+
-          echo "Cuda install location: ${{steps.cuda-toolkit.outputs.CUDA_PATH}}"
-          nvcc -V
-      - uses: actions/checkout@v5
-
-      - name: Install Packages
-        run: |
-          sudo apt-get install -y cmake make
-      - name: Setup Boost
-        run: |
-          echo GITHUB_REPOSITORY: $GITHUB_REPOSITORY
-          LIBRARY=${GITHUB_REPOSITORY#*/}
-          echo LIBRARY: $LIBRARY
-          echo "LIBRARY=$LIBRARY" >> $GITHUB_ENV
-          echo GITHUB_BASE_REF: $GITHUB_BASE_REF
-          echo GITHUB_REF: $GITHUB_REF
-          REF=${GITHUB_BASE_REF:-$GITHUB_REF}
-          REF=${REF#refs/heads/}
-          echo REF: $REF
-          BOOST_BRANCH=develop && [ "$REF" == "master" ] && BOOST_BRANCH=master || true
-          echo BOOST_BRANCH: $BOOST_BRANCH
-          cd ..
-          git clone -b $BOOST_BRANCH --depth 1 https://github.com/boostorg/boost.git boost-root
-          cd boost-root
-          mkdir -p libs/$LIBRARY
-          cp -r $GITHUB_WORKSPACE/* libs/$LIBRARY
-          git submodule update --init tools/boostdep
-          python3 tools/boostdep/depinst/depinst.py --git_args "--jobs 3" $LIBRARY
-      # ctest --output-on-failure --no-tests=error should be added between the last cmake and cd when GPU runners are available
-      - name: Test C++17/20/23
-        run: |
-          for std in 17 20 23; do
-            echo "======== Testing C++${std} ========"
-            cd ../boost-root
-            rm -rf __build__
-            mkdir __build__ && cd __build__
-            cmake -DBOOST_INCLUDE_LIBRARIES=$LIBRARY -DBUILD_TESTING=ON -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc -DBOOST_DECIMAL_ENABLE_CUDA=1 -DCMAKE_CUDA_ARCHITECTURES="75;86" -DCMAKE_CUDA_STANDARD=${std} ..
-            cmake --build . --target tests -j $(nproc)
-            cd $GITHUB_WORKSPACE
-          done
diff --git a/doc/modules/ROOT/pages/benchmarks.adoc b/doc/modules/ROOT/pages/benchmarks.adoc
index 52936c83e..488c2c22a 100644
--- a/doc/modules/ROOT/pages/benchmarks.adoc
+++ b/doc/modules/ROOT/pages/benchmarks.adoc
@@ -62,7 +62,7 @@ These benchmarks are automatically disabled if your compiler does not provide fe
 [#x64_linux_benchmarks]
 == x64 Linux
 
-Run using an Intel i9-11900k chipset running Ubuntu 24.04 and Intel oneAPI compiler 2025.2.0 or GCC 13.3.0.
+Run using an Intel i9-11900k chipset running Ubuntu 24.04 and Intel oneAPI compiler 2025.3.3 or GCC 13.4.0.
 
 === Comparisons
 
@@ -71,38 +71,38 @@ Intel Compiler:
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 72,696
-| 0.505
+| 70,801
+| 0.499
 | `double`
-| 143,924
+| 141,854
 | 1.000
 | `decimal32_t`
-| 1,485,786
-| 10.323
+| 1,322,218
+| 9.321
 | `decimal64_t`
-| 1,653,991
-| 11.492
+| 1,599,987
+| 11.279
 | `decimal128_t`
-| 4,662,704
-| 32.397
+| 3,951,400
+| 27.855
 | `decimal_fast32_t`
-| 619,662
-| 4.305
+| 537,268
+| 3.787
 | `decimal_fast64_t`
-| 606,382
-| 4.213
+| 556,455
+| 3.923
 | `decimal_fast128_t`
-| 698,945
-| 4.856
+| 667,681
+| 4.707
 | Intel `BID_UINT32`
-| 2,411,294
-| 16.754
+| 3,680,493
+| 25.946
 | Intel `BID_UINT64`
-| 3,158,422
-| 21.945
+| 5,714,482
+| 40.284
 | Intel `BID_UINT128`
-| 3,389,883
-| 23.553
+| 5,853,596
+| 41.265
 |===
 
 GCC:
@@ -110,47 +110,47 @@ GCC:
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 56,457
-| 0.916
+| 55,070
+| 0.952
 | `double`
-| 61,615
+| 57,870
 | 1.000
 | `decimal32_t`
-| 1,404,638
-| 22.797
+| 1,103,116
+| 19.062
 | `decimal64_t`
-| 1,408,074
-| 22.853
+| 1,090,026
+| 18.836
 | `decimal128_t`
-| 4,974,170
-| 80.730
+| 3,772,309
+| 65.186
 | `decimal_fast32_t`
-| 546,836
-| 8.875
+| 465,000
+| 8.035
 | `decimal_fast64_t`
-| 472,387
-| 7.667
+| 513,587
+| 8.875
 | `decimal_fast128_t`
-| 480,853
-| 7.804
+| 444,145
+| 7.675
 | GCC `_Decimal32`
-| 816,703
-| 13.255
+| 831,820
+| 14.374
 | GCC `_Decimal64`
-| 501,479
-| 8.139
+| 475,605
+| 8.219
 | GCC `_Decimal128`
-| 914,600
-| 14.844
+| 872,026
+| 15.069
 | Intel `BID_UINT32`
-| 3,718,385
-| 60.348
+| 3,658,360
+| 63.217
 | Intel `BID_UINT64`
-| 5,721,887
-| 92.865
+| 5,452,325
+| 94.217
 | Intel `BID_UINT128`
-| 7,090,648
-| 115.080
+| 5,705,342
+| 98.589
 |===
 
 === Addition
@@ -160,38 +160,38 @@ Intel Compiler:
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 118,040
-| 1.303
+| 76,312
+| 1.105
 | `double`
-| 90,579
+| 69,075
 | 1.000
 | `decimal32_t`
-| 1,712,396
-| 18.905
+| 1,238,019
+| 17.923
 | `decimal64_t`
-| 1,575,893
-| 17.398
+| 1,087,951
+| 15.750
 | `decimal128_t`
-| 3,181,562
-| 35.125
+| 2,381,956
+| 34.484
 | `decimal_fast32_t`
-| 729,257
-| 8.051
+| 944,208
+| 13.669
 | `decimal_fast64_t`
-| 1,083,923
-| 11.967
+| 754,414
+| 10.922
 | `decimal_fast128_t`
-| 1,367,004
-| 15.092
+| 767,338
+| 11.109
 | Intel `BID_UINT32`
-| 1,242,797
-| 13.721
+| 2,895,250
+| 41.915
 | Intel `BID_UINT64`
-| 1,689,585
-| 18.653
+| 3,477,761
+| 50.348
 | Intel `BID_UINT128`
-| 1,958,345
-| 21.620
+| 2,587,966
+| 37.466
 |===
 
 GCC:
@@ -199,47 +199,47 @@ GCC:
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 79,256
-| 1.085
+| 72,939
+| 1.166
 | `double`
-| 73,017
+| 62,541
 | 1.000
 | `decimal32_t`
-| 1,501,645
-| 20.566
+| 1,225,447
+| 19.594
 | `decimal64_t`
-| 1,567,250
-| 21.464
+| 1,257,515
+| 20.107
 | `decimal128_t`
-| 4,609,413
-| 63.128
+| 2,085,250
+| 33.342
 | `decimal_fast32_t`
-| 735,864
-| 10.078
+| 771,283
+| 12.332
 | `decimal_fast64_t`
-| 1,002,119
-| 13.724
+| 1,064,025
+| 17.013
 | `decimal_fast128_t`
-| 1,329,644
-| 18.210
+| 926,772
+| 14.819
 | GCC `_Decimal32`
-| 2,975,146
-| 40.746
+| 2,894,703
+| 46.285
 | GCC `_Decimal64`
-| 2,186,565
-| 29.946
+| 2,100,094
+| 33.579
 | GCC `_Decimal128`
-| 3,368,864
-| 46.138
+| 3,264,420
+| 52.196
 | Intel `BID_UINT32`
-| 2,838,194
-| 38.879
+| 2,574,559
+| 41.166
 | Intel `BID_UINT64`
-| 3,297,652
-| 45.163
+| 3,379,104
+| 54.030
 | Intel `BID_UINT128`
-| 2,796,283
-| 38.296
+| 2,509,923
+| 40.132
 |===
 
 === Subtraction
@@ -249,38 +249,38 @@ Intel Compiler:
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 78,250
-| 1.069
+| 76,299
+| 1.104
 | `double`
-| 73,193
+| 69,112
 | 1.000
 | `decimal32_t`
-| 1,480,678
-| 20.229
+| 1,274,392
+| 18.440
 | `decimal64_t`
-| 1,371,677
-| 18.741
+| 1,181,406
+| 17.094
 | `decimal128_t`
-| 2,768,955
-| 37.831
+| 2,343,671
+| 33.911
 | `decimal_fast32_t`
-| 1,040,587
-| 14.217
+| 1,003,326
+| 14.517
 | `decimal_fast64_t`
-| 1,055,980
-| 14.427
+| 823,770
+| 11.919
 | `decimal_fast128_t`
-| 1,212,405
-| 16.564
+| 945,118
+| 13.675
 | Intel `BID_UINT32`
-| 1,922,108
-| 26.261
+| 3,853,532
+| 55.758
 | Intel `BID_UINT64`
-| 1,793,879
-| 24.509
+| 3,550,643
+| 51.375
 | Intel `BID_UINT128`
-| 2,397,372
-| 32.754
+| 3,186,584
+| 46.108
 |===
 
 GCC:
@@ -288,47 +288,47 @@ GCC:
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 275,230
-| 0.936
+| 267,529
+| 0.976
 | `double`
-| 293,907
+| 274,107
 | 1.000
 | `decimal32_t`
-| 1,451,610
-| 4.939
+| 1,228,320
+| 4.481
 | `decimal64_t`
-| 1,456,587
-| 4.956
+| 1,313,266
+| 4.791
 | `decimal128_t`
-| 4,332,644
-| 14.742
+| 2,234,926
+| 8.153
 | `decimal_fast32_t`
-| 842,910
-| 2.868
+| 790,531
+| 2.884
 | `decimal_fast64_t`
-| 968,939
-| 3.297
+| 1,166,909
+| 4.257
 | `decimal_fast128_t`
-| 1,327,411
-| 4.516
+| 1,445,687
+| 5.274
 | GCC `_Decimal32`
-| 2,045,306
-| 6.959
+| 1,972,600
+| 7.196
 | GCC `_Decimal64`
-| 1,355,777
-| 4.613
+| 1,313,261
+| 4.791
 | GCC `_Decimal128`
-| 3,178,891
-| 10.816
+| 3,084,987
+| 11.255
 | Intel `BID_UINT32`
-| 3,762,566
-| 12.802
+| 3,539,123
+| 12.911
 | Intel `BID_UINT64`
-| 3,432,814
-| 11.680
+| 3,522,244
+| 12.850
 | Intel `BID_UINT128`
-| 3,725,534
-| 12.676
+| 3,042,150
+| 11.098
 |===
 
 === Multiplication
@@ -338,38 +338,38 @@ Intel Compiler:
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 78,445
-| 1.078
+| 77,440
+| 1.122
 | `double`
-| 72,798
+| 68,993
 | 1.000
 | `decimal32_t`
-| 1,735,239
-| 23.836
+| 2,142,304
+| 31.051
 | `decimal64_t`
-| 2,272,739
-| 31.220
+| 2,378,675
+| 34.477
 | `decimal128_t`
-| 6,396,750
-| 87.870
+| 6,705,642
+| 97.193
 | `decimal_fast32_t`
-| 993,256
-| 13.644
+| 1,616,807
+| 23.434
 | `decimal_fast64_t`
-| 1,670,141
-| 22.942
+| 1,934,743
+| 28.043
 | `decimal_fast128_t`
-| 5,959,977
-| 81.870
+| 5,788,000
+| 83.893
 | Intel `BID_UINT32`
-| 1,375,434
-| 18.894
+| 2,719,047
+| 39.410
 | Intel `BID_UINT64`
-| 2,052,278
-| 28.191
+| 5,078,522
+| 73.609
 | Intel `BID_UINT128`
-| 5,964,489
-| 81.932
+| 21,265,835
+| 308.232
 |===
 
 GCC:
@@ -377,47 +377,47 @@ GCC:
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 76,238
-| 1.161
+| 72,857
+| 1.165
 | `double`
-| 65,652
+| 62,531
 | 1.000
 | `decimal32_t`
-| 1,703,365
-| 25.945
+| 2,112,818
+| 33.788
 | `decimal64_t`
-| 2,564,605
-| 39.063
+| 2,528,577
+| 40.437
 | `decimal128_t`
-| 7,115,514
-| 108.382
+| 7,966,781
+| 127.405
 | `decimal_fast32_t`
-| 1,225,047
-| 18.660
+| 1,130,499
+| 18.079
 | `decimal_fast64_t`
-| 1,904,509
-| 29.009
+| 2,312,590
+| 36.983
 | `decimal_fast128_t`
-| 6,056,348
-| 92.249
+| 8,116,908
+| 129.806
 | GCC `_Decimal32`
-| 2,635,531
-| 40.144
+| 2,576,432
+| 41.202
 | GCC `_Decimal64`
-| 2,545,441
-| 38.772
+| 2,462,813
+| 39.385
 | GCC `_Decimal128`
-| 7,050,299
-| 107.289
+| 6,683,052
+| 106.876
 | Intel `BID_UINT32`
-| 2,638,999
-| 40.197
+| 2,635,154
+| 42.142
 | Intel `BID_UINT64`
-| 4,605,497
-| 70.150
+| 4,973,224
+| 79.532
 | Intel `BID_UINT128`
-| 13,075,436
-| 199.163
+| 21,352,000
+| 341.463
 |===
 
 === Division
@@ -427,38 +427,38 @@ Intel Compiler:
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 100,799
-| 0.971
+| 99,024
+| 0.943
 | `double`
-| 103,796
+| 105,042
 | 1.000
 | `decimal32_t`
-| 2,134,312
-| 20.563
+| 2,416,706
+| 23.007
 | `decimal64_t`
-| 5,399,276
-| 52.018
+| 2,516,984
+| 23.962
 | `decimal128_t`
-| 10,012,578
-| 96.464
+| 5,995,607
+| 57.078
 | `decimal_fast32_t`
-| 1,558,774
-| 15.018
+| 1,708,582
+| 16.266
 | `decimal_fast64_t`
-| 1,597,873
-| 15.394
+| 1,990,538
+| 18.950
 | `decimal_fast128_t`
-| 8,105,004
-| 78.086
+| 4,335,834
+| 41.277
 | Intel `BID_UINT32`
-| 1,561,213
-| 15.041
+| 3,600,610
+| 34.278
 | Intel `BID_UINT64`
-| 3,115,862
-| 30.019
+| 5,739,738
+| 54.642
 | Intel `BID_UINT128`
-| 7,474,712
-| 72.013
+| 15,019,593
+| 142.987
 |===
 
 GCC:
@@ -466,47 +466,47 @@ GCC:
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 60,277
-| 0.747
+| 58,777
+| 0.761
 | `double`
-| 80,676
+| 77,196
 | 1.000
 | `decimal32_t`
-| 2,396,732
-| 29.708
+| 2,079,584
+| 26.939
 | `decimal64_t`
-| 4,021,720
-| 49.850
+| 2,477,436
+| 32.093
 | `decimal128_t`
-| 10,677,625
-| 132.352
+| 5,854,061
+| 75.834
 | `decimal_fast32_t`
-| 1,083,011
-| 13.424
+| 1,718,838
+| 22.266
 | `decimal_fast64_t`
-| 1,851,520
-| 22.950
+| 1,724,535
+| 22.340
 | `decimal_fast128_t`
-| 8,121,160
-| 100.664
+| 4,442,123
+| 57.543
 | GCC `_Decimal32`
-| 5,082,812
-| 63.002
+| 4,868,534
+| 63.067
 | GCC `_Decimal64`
-| 3,005,153
-| 37.250
+| 2,904,760
+| 37.628
 | GCC `_Decimal128`
-| 10,257,437
-| 130.490
+| 9,906,049
+| 128.323
 | Intel `BID_UINT32`
-| 3,242,695
-| 40.194
+| 3,356,191
+| 43.476
 | Intel `BID_UINT64`
-| 6,143,554
-| 76.151
+| 5,745,255
+| 74.424
 | Intel `BID_UINT128`
-| 13,499,022
-| 167.324
+| 14,918,326
+| 193.253
 |===
 
 === `from_chars`
@@ -1080,36 +1080,36 @@ Run using an Intel i9-11900k chipset running Windows 11 and Visual Studio 17.14.
 [#m4_mac_benchmarks]
 == ARM64 macOS
 
-Run using a Macbook pro with M4 Max chipset running macOS Sequoia 15.5 and homebrew Clang 20.1.8
+Run using a Macbook pro with M4 Max chipset running macOS Tahoe 26.5.1 and homebrew Clang 22.1.7
 
 === Comparisons
 
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 64,639
-| 1.606
+| 67,936
+| 1.536
 | `double`
-| 40,255
+| 44,239
 | 1.000
 | `decimal32_t`
-| 957,179
-| 23.778
+| 998,053
+| 22.560
 | `decimal64_t`
-| 897,409
-| 22.293
+| 958,425
+| 21.665
 | `decimal128_t`
-| 2,131,391
-| 52.947
+| 2,055,981
+| 46.474
 | `decimal_fast32_t`
-| 380,892
-| 9.462
+| 506,597
+| 11.451
 | `decimal_fast64_t`
-| 481,455
-| 11.960
+| 473,861
+| 10.711
 | `decimal_fast128_t`
-| 465,461
-| 11.563
+| 441,387
+| 9.977
 |===
 
 === Addition
@@ -1117,29 +1117,29 @@ Run using a Macbook pro with M4 Max chipset running macOS Sequoia 15.5 and homeb
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 11,853
-| 0.964
+| 12,488
+| 0.994
 | `double`
-| 12,295
+| 12,561
 | 1.000
 | `decimal32_t`
-| 1,338,796
-| 108.889
+| 1,111,870
+| 88.518
 | `decimal64_t`
-| 1,231,462
-| 100.160
+| 825,194
+| 65.695
 | `decimal128_t`
-| 2,262,808
-| 184.043
+| 1,439,508
+| 114.601
 | `decimal_fast32_t`
-| 608,660
-| 49.505
+| 687,518
+| 54.734
 | `decimal_fast64_t`
-| 847,512
-| 68.931
+| 610,775
+| 48.625
 | `decimal_fast128_t`
-| 1,030,662
-| 83.827
+| 693,826
+| 55.237
 |===
 
 === Subtraction
@@ -1147,29 +1147,29 @@ Run using a Macbook pro with M4 Max chipset running macOS Sequoia 15.5 and homeb
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 11,939
-| 0.951
+| 12,655
+| 1.001
 | `double`
-| 12,551
+| 12,645
 | 1.000
 | `decimal32_t`
-| 1,296,430
-| 103.293
+| 1,151,474
+| 91.062
 | `decimal64_t`
-| 1,180,456
-| 94.053
+| 910,530
+| 72.007
 | `decimal128_t`
-| 2,078,008
-| 165.565
+| 1,537,309
+| 121.574
 | `decimal_fast32_t`
-| 817,989
-| 65.173
+| 755,153
+| 59.719
 | `decimal_fast64_t`
-| 823,569
-| 65.618
+| 702,725
+| 55.573
 | `decimal_fast128_t`
-| 993,447
-| 79.153
+| 789,072
+| 62.402
 |===
 
 === Multiplication
@@ -1177,29 +1177,29 @@ Run using a Macbook pro with M4 Max chipset running macOS Sequoia 15.5 and homeb
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 12,186
-| 0.944
+| 12,866
+| 1.000
 | `double`
-| 12,914
+| 12,869
 | 1.000
 | `decimal32_t`
-| 1,441,141
-| 111.595
+| 2,104,935
+| 163.566
 | `decimal64_t`
-| 2,117,061
-| 163.935
+| 2,157,113
+| 167.621
 | `decimal128_t`
-| 5,376,470
-| 416.329
+| 5,183,487
+| 402.789
 | `decimal_fast32_t`
-| 923,346
-| 71.500
+| 1,536,569
+| 119.401
 | `decimal_fast64_t`
-| 1,766,419
-| 136.783
+| 1,934,438
+| 150.318
 | `decimal_fast128_t`
-| 5,463,675
-| 423.082
+| 4,762,463
+| 370.072
 |===
 
 === Division
@@ -1207,29 +1207,29 @@ Run using a Macbook pro with M4 Max chipset running macOS Sequoia 15.5 and homeb
 |===
 | Type | Runtime (us) | Ratio to `double`
 | `float`
-| 12,576
-| 0.722
+| 13,501
+| 0.750
 | `double`
-| 17,145
+| 18,001
 | 1.000
 | `decimal32_t`
-| 1,732,611
-| 101.056
+| 2,210,859
+| 122.819
 | `decimal64_t`
-| 3,558,094
-| 207.529
+| 2,495,550
+| 138.634
 | `decimal128_t`
-| 8,985,521
-| 524.090
+| 8,639,540
+| 479.948
 | `decimal_fast32_t`
-| 1,075,184
-| 62.711
+| 1,557,445
+| 86.520
 | `decimal_fast64_t`
-| 2,027,533
-| 118.258
+| 2,394,632
+| 133.028
 | `decimal_fast128_t`
-| 7,583,016
-| 442.287
+| 7,662,160
+| 425.652
 |===
 
 === `from_chars`
@@ -1415,3 +1415,158 @@ Run using a Macbook pro with M4 Max chipset running macOS Sequoia 15.5 and homeb
 | 5,300,774
 | 1.189
 |===
+
+[#arm64_windows_benchmarks]
+== ARM64 Windows
+
+Run on the GitHub Actions `windows-11-arm` runner using MSVC in release mode.
+
+=== Comparisons
+
+|===
+| Type | Runtime (us) | Ratio to `double`
+| `float`
+| 182,667
+| 1.012
+| `double`
+| 180,555
+| 1.000
+| `decimal32_t`
+| 1,679,312
+| 9.301
+| `decimal64_t`
+| 2,203,320
+| 12.203
+| `decimal128_t`
+| 8,833,555
+| 48.924
+| `decimal_fast32_t`
+| 872,493
+| 4.832
+| `decimal_fast64_t`
+| 891,568
+| 4.938
+| `decimal_fast128_t`
+| 907,563
+| 5.027
+|===
+
+=== Addition
+
+|===
+| Type | Runtime (us) | Ratio to `double`
+| `float`
+| 60,279
+| 0.941
+| `double`
+| 64,051
+| 1.000
+| `decimal32_t`
+| 2,209,244
+| 34.492
+| `decimal64_t`
+| 2,231,636
+| 34.842
+| `decimal128_t`
+| 5,480,362
+| 85.562
+| `decimal_fast32_t`
+| 1,502,684
+| 23.461
+| `decimal_fast64_t`
+| 1,982,389
+| 30.950
+| `decimal_fast128_t`
+| 2,889,333
+| 45.110
+|===
+
+=== Subtraction
+
+|===
+| Type | Runtime (us) | Ratio to `double`
+| `float`
+| 55,587
+| 0.918
+| `double`
+| 60,552
+| 1.000
+| `decimal32_t`
+| 2,346,070
+| 38.745
+| `decimal64_t`
+| 2,458,995
+| 40.610
+| `decimal128_t`
+| 5,268,604
+| 87.010
+| `decimal_fast32_t`
+| 1,879,880
+| 31.046
+| `decimal_fast64_t`
+| 2,346,222
+| 38.747
+| `decimal_fast128_t`
+| 3,237,777
+| 53.471
+|===
+
+=== Multiplication
+
+|===
+| Type | Runtime (us) | Ratio to `double`
+| `float`
+| 56,447
+| 0.914
+| `double`
+| 61,774
+| 1.000
+| `decimal32_t`
+| 3,276,537
+| 53.041
+| `decimal64_t`
+| 4,325,985
+| 70.029
+| `decimal128_t`
+| 15,726,575
+| 254.582
+| `decimal_fast32_t`
+| 2,313,991
+| 37.459
+| `decimal_fast64_t`
+| 3,821,815
+| 61.868
+| `decimal_fast128_t`
+| 15,496,341
+| 250.855
+|===
+
+=== Division
+
+|===
+| Type | Runtime (us) | Ratio to `double`
+| `float`
+| 56,701
+| 0.381
+| `double`
+| 149,002
+| 1.000
+| `decimal32_t`
+| 3,664,152
+| 24.591
+| `decimal64_t`
+| 7,224,334
+| 48.485
+| `decimal128_t`
+| 16,818,955
+| 112.877
+| `decimal_fast32_t`
+| 2,750,313
+| 18.458
+| `decimal_fast64_t`
+| 6,758,148
+| 45.356
+| `decimal_fast128_t`
+| 15,086,240
+| 101.249
+|===