diff --git a/mold-2.34.1.tar.gz b/mold-2.34.1.tar.gz
deleted file mode 100644
index 09d990c..0000000
--- a/mold-2.34.1.tar.gz
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:a8cf638045b4a4b2697d0bcc77fd96eae93d54d57ad3021bf03b0333a727a59d
-size 10057683
diff --git a/mold-2.35.0.tar.gz b/mold-2.35.0.tar.gz
new file mode 100644
index 0000000..9be989d
--- /dev/null
+++ b/mold-2.35.0.tar.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2703f1c88c588523815886478950bcae1ef02190dc4787e0d120a293b1a46e3b
+size 10056856
diff --git a/mold.changes b/mold.changes
index e55f5f7..b314660 100644
--- a/mold.changes
+++ b/mold.changes
@@ -1,3 +1,37 @@
+-------------------------------------------------------------------
+Sun Dec  8 08:58:41 UTC 2024 - Martin Liška <martin.liska@hey.com>
+
+- Update to version 2.35.0
+  * Big-endian ARM64 is now supported. ARM64 is a bi-endian processor, meaning that
+    the processor can run in either little- or big-endian mode. Even though
+    little-endian is the de facto standard, the ARM64 processor-specific ABI defines
+    its big-endian variant, and the ARM toolchain supports it. Now we support it
+    too. (882e7eb)
+  * Big-endian SH4 is now supported. SH4 has become a minor CPU nowadays, and its
+    big-endian variant is even more so, but some SHARP scientific calculators still
+    use SH4 processors in big-endian mode. (0cb9fc6)
+  * mold attempts to overwrite an existing file if a specified output file already
+    exists because reusing an existing file is much faster than creating a fresh
+    file and writing to it on Linux. If an existing file is currently running,
+    open(2) for that file fails with ETXTBSY. When that happens, mold falls back to
+    creating a new file. The problem here is that Linux kernel version 6.11 changed
+    that well-known behavior of open(2), and it now allows user programs to
+    overwrite a running executable. That caused a very mysterious issue for programs
+    that rebuild themselves during the build, such as gcc or ninja (#1361). Even
+    though the kernel's change has been reverted (torvalds/linux@3b83203), we need
+    to make adjustments to mold for that particular version of the Linux kernel. So,
+    if mold detects that it is running on Linux 6.11, it no longer tries to reuse an
+    existing output file. (8e4f7b5)
+  * On rare occasions, mold could fail with a "ConcurrentMap is full" error. Now the
+    issue has been resolved. (e56b649)
+  * Even if a user choose not to use mimalloc memory allocator (i.e. built mold with
+    -DMOLD_USE_MIMALLOC=0), mold was still being built with mimalloc. This issue has
+    been resolved. (ffd10dd)
+  * [s390x] s390x uses nonstandard 8-byte entries for the .hash section. Previously,
+    mold created 4-byte entries for .hash, which caused mold-generated executables
+    to crash on startup if they were built with -Wl,--hash-style=sysv. Now, mold
+    generates a psABI-compliant .hash section. (e2e1146)
+
 -------------------------------------------------------------------
 Fri Oct  4 18:37:30 UTC 2024 - Martin Liška <martin.liska@hey.com>
 
diff --git a/mold.spec b/mold.spec
index 22ad08e..8f27da9 100644
--- a/mold.spec
+++ b/mold.spec
@@ -17,7 +17,7 @@
 
 
 Name:           mold
-Version:        2.34.1
+Version:        2.35.0
 Release:        0
 Summary:        A Modern Linker (mold)
 License:        MIT
@@ -110,6 +110,5 @@ fi
 %{_mandir}/man1/ld.mold.1.gz
 %dir %{_docdir}/mold
 %doc %{_docdir}/mold/LICENSE
-%doc %{_docdir}/mold/LICENSE.third-party
 
 %changelog