diff --git a/.github/workflows/benchmark-master.yaml b/.github/workflows/benchmark-master.yaml
index dae38f30c7..0764ed7db4 100644
--- a/.github/workflows/benchmark-master.yaml
+++ b/.github/workflows/benchmark-master.yaml
@@ -26,4 +26,4 @@ jobs:
           --branch master \
           --testbed ubuntu-latest \
           --adapter rust_criterion \
-          nix develop --command cargo bench --package nickel-lang-core --bench numeric
+          nix develop --command cargo bench --package nickel-lang-core --features=benchmark-ci --bench numeric --bench arrays
diff --git a/.github/workflows/benchmark-pr.yaml b/.github/workflows/benchmark-pr.yaml
index 973f918caa..92db0cc2af 100644
--- a/.github/workflows/benchmark-pr.yaml
+++ b/.github/workflows/benchmark-pr.yaml
@@ -31,7 +31,7 @@ jobs:
           nix_path: "nixpkgs=channel:nixos-unstable"
       - name: run benchmarks
         run: |
-          nix develop --command cargo bench --package nickel-lang-core --bench numeric > criterion-output.txt
+          nix develop --command cargo bench --package nickel-lang-core --features=benchmark-ci --bench numeric --bench arrays > criterion-output.txt
       - name: upload results
         uses: actions/upload-artifact@v4
         with:
diff --git a/core/Cargo.toml b/core/Cargo.toml
index 1661c2528c..3b91f85394 100644
--- a/core/Cargo.toml
+++ b/core/Cargo.toml
@@ -28,6 +28,7 @@ format = ["dep:topiary-core", "dep:topiary-queries", "dep:tree-sitter-nickel"]
 metrics = ["dep:metrics"]
 nix-experimental = [ "dep:cxx", "dep:cxx-build", "dep:pkg-config" ]
 spanned-deser = ["dep:serde-untagged"]
+benchmark-ci = []
 
 [build-dependencies]
 lalrpop.workspace = true
diff --git a/core/benches/arrays.rs b/core/benches/arrays.rs
index 9c8d36bf83..6634f48ec5 100644
--- a/core/benches/arrays.rs
+++ b/core/benches/arrays.rs
@@ -1,3 +1,5 @@
+#![cfg_attr(feature = "benchmark-ci", allow(unused_imports))]
+
 use std::rc::Rc;
 
 use criterion::criterion_main;
@@ -9,6 +11,7 @@ use nickel_lang_utils::{bench::criterion_config, bench::EvalMode, ncl_bench_grou
 use pretty::{BoxAllocator, DocBuilder, Pretty};
 
 /// Generates a pseaudo-random Nickel array as a string.
+#[cfg(not(feature = "benchmark-ci"))]
 fn ncl_random_array(len: usize) -> String {
     let m = 2_u64.pow(32);
     let a = 1664525;
@@ -32,6 +35,7 @@ fn ncl_random_array(len: usize) -> String {
     String::from_utf8(out).unwrap()
 }
 
+#[cfg(not(feature = "benchmark-ci"))]
 ncl_bench_group! {
 name = benches;
 config = criterion_config();
@@ -168,4 +172,63 @@ config = criterion_config();
         eval_mode = EvalMode::DeepSeq,
     }
 }
+
+#[cfg(feature = "benchmark-ci")]
+ncl_bench_group! {
+name = benches;
+config = criterion_config();
+{
+        name = "foldr strings 50",
+        path = "arrays/fold",
+        subtest = "right.strings",
+        args = (50),
+    }, {
+        name = "foldr strings 500",
+        path = "arrays/fold",
+        subtest = "right.strings",
+        args = (500),
+    }, {
+        name = "foldl arrays 50",
+        path = "arrays/fold",
+        subtest = "left.arrays",
+        args = (50),
+    }, {
+        name = "foldl arrays 500",
+        path = "arrays/fold",
+        subtest = "left.arrays",
+        args = (500),
+    }, {
+        name = "generate normal 50",
+        path = "arrays/generate",
+        subtest = "checked",
+        args = (50),
+    }, {
+        name = "generate normal 250",
+        path = "arrays/generate",
+        subtest = "checked",
+        // Most other benchmarks have a factor of 10 between
+        // the small and large sizes, but this one is slow so
+        // use a factor of 5.
+        args = (250),
+    }, {
+        name = "generate normal unchecked 200",
+        path = "arrays/generate",
+        subtest = "unchecked",
+        args = (200),
+    }, {
+        name = "generate normal unchecked 1000",
+        path = "arrays/generate",
+        subtest = "unchecked",
+        args = (1000),
+    }, {
+        name = "pipe normal 20",
+        path = "arrays/pipe",
+        args = (20),
+    }, {
+        name = "pipe normal 200",
+        path = "arrays/pipe",
+        args = (200),
+    },
+}
+
 criterion_main!(benches);
diff --git a/core/benches/arrays/sum.ncl b/core/benches/arrays/sum.ncl
index 7b66cc5887..180ead325b 100644
--- a/core/benches/arrays/sum.ncl
+++ b/core/benches/arrays/sum.ncl
@@ -1,9 +1,8 @@
 let rec sum | Array Number -> Number
-  = fun xs =>
-    if std.array.length xs == 0 then
-      0
-    else
-      std.array.first xs + sum (std.array.drop_first xs)
+  = match {
+    [] => 0,
+    [x, ..xs] => x + sum xs,
+  }
   in
 {
   run = fun n => std.array.generate (fun x => x + 1) n |> sum