diff --git a/Cargo.lock b/Cargo.lock
index 4e9b54bb8aa..470f9bc919b 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -429,15 +429,6 @@ dependencies = [
  "libbz2-rs-sys",
 ]
 
-[[package]]
-name = "caseless"
-version = "0.2.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8b6fd507454086c8edfd769ca6ada439193cdb209c7681712ef6275cccbfe5d8"
-dependencies = [
- "unicode-normalization",
-]
-
 [[package]]
 name = "cast"
 version = "0.3.0"
@@ -3561,7 +3552,6 @@ dependencies = [
  "ascii",
  "bitflags 2.11.0",
  "bstr",
- "caseless",
  "chrono",
  "constant_time_eq",
  "crossbeam-utils",
diff --git a/Cargo.toml b/Cargo.toml
index 80976b1e1c7..4ba2fef2e48 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -186,7 +186,6 @@ bitflags = "2.11.0"
 bitflagset = "0.0.3"
 bstr = "1"
 bzip2 = "0.6"
-caseless = "0.2.2"
 chrono = { version = "0.4.44", default-features = false, features = ["clock", "std"] }
 console_error_panic_hook = "0.1"
 constant_time_eq = "0.4"
diff --git a/crates/vm/Cargo.toml b/crates/vm/Cargo.toml
index 869f3418f11..e04e65fce2d 100644
--- a/crates/vm/Cargo.toml
+++ b/crates/vm/Cargo.toml
@@ -73,7 +73,6 @@ strum_macros = { workspace = true }
 thiserror = { workspace = true }
 memchr = { workspace = true }
 
-caseless = { workspace = true }
 flamer = { workspace = true, optional = true }
 half = { workspace = true }
 psm = { workspace = true }
diff --git a/crates/vm/src/builtins/str.rs b/crates/vm/src/builtins/str.rs
index 960c3581301..212b14c7487 100644
--- a/crates/vm/src/builtins/str.rs
+++ b/crates/vm/src/builtins/str.rs
@@ -45,10 +45,10 @@ use rustpython_common::{
     hash,
     lock::PyMutex,
     str::DeduceStrKind,
-    wtf8::{CodePoint, Wtf8, Wtf8Buf, Wtf8Chunk, Wtf8Concat},
+    wtf8::{CodePoint, Wtf8, Wtf8Buf, Wtf8Concat},
 };
 
-use icu_casemap::TitlecaseMapper;
+use icu_casemap::{CaseMapper, TitlecaseMapper};
 use icu_locale::LanguageIdentifier;
 use icu_properties::props::{
     BidiClass, BinaryProperty, CaseIgnorable, Cased, EnumeratedProperty, GeneralCategory,
@@ -743,20 +743,31 @@ impl PyStr {
         }
     }
 
-    // casefold is much more aggressive than lower
+    // Case folding is a Unicode standard operation to erase case differences.
+    //
+    // Lower, upper, and title case are special properties. Case folding erases those
+    // differences. For ASCII, case folding is the same as lower case but other scripts have
+    // their own, well-defined mappings.
     #[pymethod]
     fn casefold(&self) -> Self {
         match self.as_str_kind() {
-            PyKindStr::Ascii(s) => caseless::default_case_fold_str(s.as_str()).into(),
-            PyKindStr::Utf8(s) => caseless::default_case_fold_str(s).into(),
-            PyKindStr::Wtf8(w) => w
-                .chunks()
-                .map(|c| match c {
-                    Wtf8Chunk::Utf8(s) => Wtf8Buf::from_string(caseless::default_case_fold_str(s)),
-                    Wtf8Chunk::Surrogate(c) => Wtf8Buf::from(c),
-                })
-                .collect::<Wtf8Buf>()
-                .into(),
+            PyKindStr::Ascii(s) => s.to_ascii_lowercase().into(),
+            PyKindStr::Utf8(s) => CaseMapper::new().fold_string(s).to_string().into(),
+            PyKindStr::Wtf8(w) => {
+                let mut out = VecFmtWriter(Vec::with_capacity(w.len()));
+                let mapper = CaseMapper::new();
+                for chunk in w.as_bytes().utf8_chunks() {
+                    mapper
+                        .fold(chunk.valid())
+                        .write_to(&mut out)
+                        .expect("Writing to an in-memory buffer cannot fail.");
+                    out.0.extend(chunk.invalid());
+                }
+                // SAFETY:
+                // * CaseMapper only produces valid UTF-8
+                // * Surrogates are appended as-is
+                unsafe { Wtf8Buf::from_bytes_unchecked(out.0) }.into()
+            }
         }
     }