diff --git a/.github/workflows/dev-long-tests.yml b/.github/workflows/dev-long-tests.yml
index 4b0f6c2acee..fab0b96203f 100644
--- a/.github/workflows/dev-long-tests.yml
+++ b/.github/workflows/dev-long-tests.yml
@@ -30,7 +30,7 @@ jobs:
     - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # tag=v7.0.1
     - name: make test
       run: |
-        make test
+        make -j test
         make -j zstd
         ./tests/test_process_substitution.bash ./zstd
 
@@ -40,7 +40,7 @@ jobs:
     steps:
     - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # tag=v7.0.1
     - name: make test on macos
-      run: make test
+      run: make -j test
 
   # lasts ~10mn
   make-test-32bit:
@@ -72,7 +72,7 @@ jobs:
     steps:
     - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # tag=v7.0.1
     - name: no intrinsics fuzztest
-      run: MOREFLAGS="-DZSTD_NO_INTRINSICS" make -C tests fuzztest
+      run: MOREFLAGS="-DZSTD_NO_INTRINSICS" make -j -C tests fuzztest
 
   # lasts ~8mn
   tsan-zstreamtest:
@@ -95,7 +95,7 @@ jobs:
     steps:
     - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # tag=v7.0.1
     - name: thread sanitizer fuzztest
-      run: CC=clang make tsan-fuzztest
+      run: CC=clang make -j tsan-fuzztest
 
   big-tests-zstreamtest32:
     runs-on: ubuntu-latest
@@ -151,14 +151,14 @@ jobs:
         echo "deb [arch=amd64] http://archive.ubuntu.com/ubuntu focal main universe" | sudo tee -a /etc/apt/sources.list
         sudo apt-get -qqq update
         make gcc8install
-        CC=gcc-8 FUZZER_FLAGS="--long-tests" make clean uasan-fuzztest
+        CC=gcc-8 FUZZER_FLAGS="--long-tests" make -j clean uasan-fuzztest
 
   clang-asan-ubsan-fuzz:
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # tag=v7.0.1
     - name: clang + ASan + UBSan + Fuzz Test
-      run: CC=clang FUZZER_FLAGS="--long-tests" make clean uasan-fuzztest
+      run: CC=clang FUZZER_FLAGS="--long-tests" make -j clean uasan-fuzztest
 
   # the only job decoding real legacy frames under sanitizers.
   # fuzzers do build these decoders, but only reach them if an input
@@ -178,7 +178,7 @@ jobs:
       run: |
         sudo apt-get -qqq update
         make libc6install
-        CFLAGS="-O3 -m32" FUZZER_FLAGS="--long-tests" make uasan-fuzztest
+        CFLAGS="-O3 -m32" FUZZER_FLAGS="--long-tests" make -j uasan-fuzztest
 
   clang-asan-fuzz32:
     runs-on: ubuntu-latest
@@ -188,7 +188,7 @@ jobs:
       run: |
         sudo apt-get -qqq update
         make libc6install
-        CC=clang CFLAGS="-O3 -m32" FUZZER_FLAGS="--long-tests" make asan-fuzztest
+        CC=clang CFLAGS="-O3 -m32" FUZZER_FLAGS="--long-tests" make -j asan-fuzztest
 
 # The following test seems to have issues on github CI specifically,
 # it does not provide the `__mulodi4` instruction emulation
@@ -231,10 +231,12 @@ jobs:
     steps:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # tag=v7.0.1
       - name: clang + MSan + Fuzz Test
+        # -O0 targets the unit tests on unoptimized code: random fuzzing is kept short.
+        # clang-msan-fuzz runs the long random fuzzing.
         run: |
           sudo apt-get -qqq update
           sudo apt-get install clang
-          CC=clang MOREFLAGS="-O0" make clean msan-fuzztest
+          CC=clang MOREFLAGS="-O0" FUZZERTEST=-T60s ZSTREAM_TESTTIME=-T30s make clean msan-fuzztest
 
   clang-msan-fuzz:
     runs-on: ubuntu-latest
@@ -251,8 +253,10 @@ jobs:
     steps:
     - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # tag=v7.0.1
     - name: clang + MSan + Test Zstd
+      # ZSTDRTTEST= : skip the large data tests of playTests.sh, too slow under MSan.
+      # They still run in the make-test* and ASan + UBSan testzstd jobs.
       run: |
-        CC=clang make -j msan-test-zstd HAVE_ZLIB=0 HAVE_LZ4=0 HAVE_LZMA=0 V=1
+        CC=clang make -j msan-test-zstd HAVE_ZLIB=0 HAVE_LZ4=0 HAVE_LZMA=0 ZSTDRTTEST= V=1
 
   armfuzz:
       runs-on: ubuntu-latest
diff --git a/.gitignore b/.gitignore
index 03ecbde6af8..6868bc7245d 100644
--- a/.gitignore
+++ b/.gitignore
@@ -58,6 +58,7 @@ googletest/
 *.d
 *.vscode
 *.code-workspace
+.claude/
 compile_commands.json
 .clangd
 perf.data
diff --git a/tests/fuzz/README.md b/tests/fuzz/README.md
index e2196e832c6..8732c8c3978 100644
--- a/tests/fuzz/README.md
+++ b/tests/fuzz/README.md
@@ -105,7 +105,7 @@ CC=afl-clang CXX=afl-clang++ ./fuzz.py build all --enable-asan --enable-ubsan
 ## Regression Testing
 
 The regression test supports the `all` target to run all the fuzzers in one
-command.
+command. Targets run in parallel, one per CPU.
 
 ```
 CC=clang CXX=clang++ ./fuzz.py build all --enable-asan --enable-ubsan
diff --git a/tests/fuzz/fuzz.py b/tests/fuzz/fuzz.py
index 492be3486d0..22a1ee3f925 100755
--- a/tests/fuzz/fuzz.py
+++ b/tests/fuzz/fuzz.py
@@ -11,6 +11,7 @@
 # ##########################################################################
 
 import argparse
+import concurrent.futures
 import contextlib
 import os
 import re
@@ -648,6 +649,13 @@ def afl(args):
     return 0
 
 
+def corpus_size(target):
+    size = 0
+    for root, _, files in os.walk(abs_join(CORPORA_DIR, target)):
+        size += sum(os.path.getsize(os.path.join(root, f)) for f in files)
+    return size
+
+
 def regression(args):
     try:
         description = """
@@ -655,18 +663,34 @@ def regression(args):
         The fuzzer should have been built with
         LIB_FUZZING_ENGINE='libregression.a'.
         Takes input from CORPORA.
+        Targets run in parallel, one per CPU.
         """
         args = targets_parser(args, description)
     except Exception as e:
         print(e)
         return 1
-    for target in args.TARGET:
+
+    def run(target):
         corpora = create(abs_join(CORPORA_DIR, target))
-        target = abs_join(FUZZ_DIR, target)
-        cmd = [target, corpora]
-        print(' '.join(cmd))
-        subprocess.check_call(cmd)
-    return 0
+        cmd = [abs_join(FUZZ_DIR, target), corpora]
+        result = subprocess.run(cmd, stdout=subprocess.PIPE,
+                                stderr=subprocess.STDOUT)
+        return cmd, result
+
+    # Largest corpora take longest: start them first
+    targets = sorted(args.TARGET, key=corpus_size, reverse=True)
+    failed = []
+    with concurrent.futures.ThreadPoolExecutor(os.cpu_count()) as pool:
+        runs = [pool.submit(run, target) for target in targets]
+        for done in concurrent.futures.as_completed(runs):
+            cmd, result = done.result()
+            print(' '.join(cmd))
+            print(result.stdout.decode(errors='replace'), end='', flush=True)
+            if result.returncode != 0:
+                failed.append(cmd[0])
+    for target in failed:
+        print('FAILED: {}'.format(target))
+    return 1 if failed else 0
 
 
 def gen_parser(args):
diff --git a/tests/playTests.sh b/tests/playTests.sh
index d8ea0c3dad3..44bc82b5610 100755
--- a/tests/playTests.sh
+++ b/tests/playTests.sh
@@ -1909,10 +1909,10 @@ println "- Create second (different) dictionary"
 zstd -T0 --train-cover=k=56,d=8 "$TESTDIR"/*.c "$PRGDIR"/*.c "$PRGDIR"/*.h -o tmpDictC
 zstd -d tmp.zst -D tmpDictC -fo result && die "wrong dictionary not detected!"
 println "- Create dictionary using shrink-dict flag"
-zstd -T0 --train-cover=steps=256,shrink "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict
-zstd -T0 --train-cover=steps=256,shrink=1 "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict1
-zstd -T0 --train-cover=steps=256,shrink=5 "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict2
-zstd -T0 --train-cover=shrink=5,steps=256 "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict3
+zstd -T0 --train-cover=steps=7,shrink "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict
+zstd -T0 --train-cover=steps=7,shrink=1 "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict1
+zstd -T0 --train-cover=steps=7,shrink=5 "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict2
+zstd -T0 --train-cover=shrink=5,steps=7 "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict3
 println "- Create dictionary with short dictID"
 zstd -T0 --train-cover=k=46,d=8,split=80 "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpDict1
 cmp tmpDict tmpDict1 && die "dictionaries should have different ID !"