diff --git a/src/Compilation.zig b/src/Compilation.zig
index 73185a91ea..f83bac1e19 100644
--- a/src/Compilation.zig
+++ b/src/Compilation.zig
@@ -5371,10 +5371,12 @@ pub fn addCCArgs(
                 try argv.append(include_dir);
             }
 
-            if (target.cpu.model.llvm_name) |llvm_name| {
-                try argv.appendSlice(&[_][]const u8{
-                    "-Xclang", "-target-cpu", "-Xclang", llvm_name,
-                });
+            if (target_util.clangSupportsTargetCpuArg(target)) {
+                if (target.cpu.model.llvm_name) |llvm_name| {
+                    try argv.appendSlice(&[_][]const u8{
+                        "-Xclang", "-target-cpu", "-Xclang", llvm_name,
+                    });
+                }
             }
 
             // It would be really nice if there was a more compact way to communicate this info to Clang.
diff --git a/src/target.zig b/src/target.zig
index 1f8a567f03..f0ef89d4e9 100644
--- a/src/target.zig
+++ b/src/target.zig
@@ -315,6 +315,20 @@ pub fn clangAssemblerSupportsMcpuArg(target: std.Target) bool {
     };
 }
 
+/// Some experimental or poorly-maintained LLVM targets do not properly process CPU models in their
+/// Clang driver code. For these, we should omit the `-Xclang -target-cpu -Xclang <model>` flags.
+pub fn clangSupportsTargetCpuArg(target: std.Target) bool {
+    return switch (target.cpu.arch) {
+        .arc,
+        .msp430,
+        .ve,
+        .xcore,
+        .xtensa,
+        => false,
+        else => true,
+    };
+}
+
 pub fn clangSupportsFloatAbiArg(target: std.Target) bool {
     return switch (target.cpu.arch) {
         .arm,