diff --git a/changelog.d/10176-regex-no-work-cap.md b/changelog.d/10176-regex-no-work-cap.md
new file mode 100644
index 0000000000..3c10963c86
--- /dev/null
+++ b/changelog.d/10176-regex-no-work-cap.md
@@ -0,0 +1,3 @@
+### Fixed
+
+- **RegExp operations are no longer capped by work** (#10164). Valid programs that Node completes threw `RangeError: Regular expression work limit exceeded`: a 32,000-unit non-ASCII `split`, a 60,000-unit global `replace`, and linear splits and replaces of 11–15 million units. No finite allowance separates valid programs from pathological ones, so a catastrophic pattern now runs as long as it does in Node. Matching still yields to the collector and to cancellation, and the memory limits are unchanged.
diff --git a/crates/perry-runtime/src/gc/tests/runtime_roots.rs b/crates/perry-runtime/src/gc/tests/runtime_roots.rs
index d21d61fe5c..a5e4bf025f 100644
--- a/crates/perry-runtime/src/gc/tests/runtime_roots.rs
+++ b/crates/perry-runtime/src/gc/tests/runtime_roots.rs
@@ -42,6 +42,8 @@ mod perex_reuse;
 mod perex_split;
 #[cfg(feature = "regex-engine")]
 mod perex_strings;
+#[cfg(feature = "regex-engine")]
+mod perex_work_policy;
 mod prototype_addr_cache;
 mod regexp_last_index;
 mod segment_record_keys;
diff --git a/crates/perry-runtime/src/gc/tests/runtime_roots/perex_work_policy.rs b/crates/perry-runtime/src/gc/tests/runtime_roots/perex_work_policy.rs
new file mode 100644
index 0000000000..dabc3f28c1
--- /dev/null
+++ b/crates/perry-runtime/src/gc/tests/runtime_roots/perex_work_policy.rs
@@ -0,0 +1,62 @@
+//! RegExp operations are not capped by work (#10164): a valid program whose
+//! matching charges more than the former 100,000,000-unit allowance completes.
+use super::*;
+use crate::regex::perex_api as api;
+use crate::regex::perex_memory::MemoryBudget;
+use crate::regex::RegExpHeader;
+use crate::string::StringHeader;
+use perex::Budget;
+
+const FORMER_WORK_LIMIT: usize = 100_000_000;
+
+fn text<'s>(scope: &'s RuntimeHandleScope, bytes: &[u8]) -> RuntimeHandle<'s> {
+    scope.root_string_ptr(crate::string::js_string_from_bytes(
+        bytes.as_ptr(),
+        bytes.len() as u32,
+    ))
+}
+
+#[test]
+fn perex_operation_allowance_admits_linear_work_beyond_the_former_limit() {
+    let _guard = CopyingNurseryTestGuard::new(0);
+    let _triggers = GcTriggerThresholdTestGuard::suppress_automatic_triggers();
+    super::perex_public::register_host_roots();
+    let scope = RuntimeHandleScope::new();
+    // One search that is linear but heavy: at every position the lookahead
+    // reads 32 units and then fails on a character class the subject never
+    // contains, so it matches nothing and allocates no results. The pattern has
+    // no required literal on purpose; a literal lets admission reject the whole
+    // search in one pass without doing the per-position work. Measured at about
+    // 220 work units per subject unit, so 540,000 units charge about 1.2e8.
+    let pattern = text(&scope, br"\w(?=[\w,; ]{32}[^\w,; ])");
+    let flags = text(&scope, b"");
+    let receiver = scope.root_raw_mut_ptr(pattern.with_const_ptr::<StringHeader, _>(|pattern| {
+        flags.with_const_ptr::<StringHeader, _>(|flags| crate::regex::js_regexp_new(pattern, flags))
+    }));
+    let input = text(&scope, "ab12,cd345;ef6 ".repeat(36_000).as_bytes());
+    let memory = MemoryBudget::new(api::SCRATCH_BYTES);
+    let mut budget = Budget::new(api::WORK);
+    let found = receiver.with_mut_ptr::<RegExpHeader, _>(|receiver| {
+        input.with_const_ptr::<StringHeader, _>(|input| {
+            api::execute_with_resources(
+                receiver,
+                input,
+                false,
+                &mut budget,
+                &memory,
+                &mut || Ok(()),
+                None,
+            )
+        })
+    });
+    let charged = api::WORK - budget.remaining();
+    assert!(
+        matches!(found, Ok(None)),
+        "a valid search must complete without a work-limit error"
+    );
+    assert!(
+        charged > FORMER_WORK_LIMIT,
+        "the witness must charge more than the former limit to prove anything; \
+         charged {charged}, so a Perex change may have made this pattern cheaper"
+    );
+}
diff --git a/crates/perry-runtime/src/regex/perex_api.rs b/crates/perry-runtime/src/regex/perex_api.rs
index 900e603280..2e23d56d6d 100644
--- a/crates/perry-runtime/src/regex/perex_api.rs
+++ b/crates/perry-runtime/src/regex/perex_api.rs
@@ -12,7 +12,18 @@ use perex::executor::ExecError;
 use perex::{span::Span, Budget};
 
 // One explicit host policy; no retained scratch cache or alternate engine.
-pub(crate) const WORK: usize = 100_000_000;
+/// A RegExp operation's work allowance: effectively unlimited (#10164).
+///
+/// JavaScript engines never abort regex matching for doing too much work, and
+/// no finite allowance separates valid programs from pathological ones: Perex's
+/// charge per subject unit depends on the program (about 1 for `/x/`, 60 for
+/// `/([a-z]+)([0-9]+)/g`, over 200 for a 32-unit lookahead), so any cap throws
+/// on some large linear input Node completes. The former 100,000,000 did, as
+/// `RangeError: Regular expression work limit exceeded`. Searches still run in
+/// `QUANTUM` slices with a GC poll between them, so collection and cancellation
+/// keep working; a catastrophic pattern runs as long as it does in Node. The
+/// memory limits below are unchanged.
+pub(crate) const WORK: usize = usize::MAX;
 pub(crate) const SCRATCH_BYTES: usize = 64 * 1024 * 1024;
 pub(crate) const PROGRAM_BYTES: usize = 32 * 1024 * 1024;
 pub(crate) const QUANTUM: usize = 4096;