diff --git a/bench_test.go b/bench_test.go
new file mode 100644
index 0000000..0504c37
--- /dev/null
+++ b/bench_test.go
@@ -0,0 +1,129 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: MIT
+
+package interpres
+
+import (
+	"fmt"
+	"strings"
+	"testing"
+	"time"
+)
+
+// benchDoc is a representative configuration document: every scalar kind, an
+// inline table, sub-tables, and an array of tables.
+var benchDoc = []byte(`title = "benchmark configuration"
+replicas = 3
+ratio = 0.75
+enabled = true
+when = 2026-09-17T12:00:00Z
+local = 2026-09-17T12:00:00
+tags = ["alpha", "beta", "gamma"]
+limits = { cpu = 4, memory = 1024 }
+
+[server]
+host = "localhost"
+port = 8080
+hosts = ["a.example", "b.example"]
+
+[server.tls]
+enabled = true
+cert = "/etc/cert.pem"
+
+[[items]]
+name = "first"
+weight = 10
+flags = ["x", "y"]
+
+[[items]]
+name = "second"
+weight = 20
+flags = ["z"]
+`)
+
+// longDoc is generated once so the large-input benchmarks measure parsing,
+// not document construction. Roughly 2000 array-of-tables entries.
+var longDoc = func() []byte {
+	var b strings.Builder
+	b.WriteString("title = \"long\"\n")
+	for i := range 2000 {
+		fmt.Fprintf(&b, "[[entry]]\nname = \"entry-%d\"\nweight = %d\nwhen = 2026-09-17T12:00:00Z\nratio = 0.5\ntags = [\"a\", \"b\", \"c\"]\n\n", i, i)
+	}
+	return []byte(b.String())
+}()
+
+type benchTLS struct {
+	Enabled bool   `toml:"enabled"`
+	Cert    string `toml:"cert"`
+}
+
+type benchServer struct {
+	Host  string   `toml:"host"`
+	Port  int      `toml:"port"`
+	Hosts []string `toml:"hosts"`
+	TLS   benchTLS `toml:"tls"`
+}
+
+type benchItem struct {
+	Name   string   `toml:"name"`
+	Weight int      `toml:"weight"`
+	Flags  []string `toml:"flags"`
+}
+
+type benchConfig struct {
+	Title    string         `toml:"title"`
+	Replicas int            `toml:"replicas"`
+	Ratio    float64        `toml:"ratio"`
+	Enabled  bool           `toml:"enabled"`
+	When     time.Time      `toml:"when"`
+	Local    LocalDateTime  `toml:"local"`
+	Tags     []string       `toml:"tags"`
+	Limits   map[string]any `toml:"limits"`
+	Server   benchServer    `toml:"server"`
+	Items    []benchItem    `toml:"items"`
+}
+
+func BenchmarkParse(b *testing.B) {
+	b.ReportAllocs()
+	b.SetBytes(int64(len(benchDoc)))
+	for b.Loop() {
+		if _, err := Parse(benchDoc); err != nil {
+			b.Fatal(err)
+		}
+	}
+}
+
+func BenchmarkMarshal(b *testing.B) {
+	tree, err := Parse(benchDoc)
+	if err != nil {
+		b.Fatal(err)
+	}
+	b.ReportAllocs()
+	b.SetBytes(int64(len(benchDoc)))
+	for b.Loop() {
+		if _, err := Marshal(tree); err != nil {
+			b.Fatal(err)
+		}
+	}
+}
+
+func BenchmarkStrictDecode(b *testing.B) {
+	dec := NewDecoder().DisallowUnknownFields()
+	b.ReportAllocs()
+	for b.Loop() {
+		var cfg benchConfig
+		if err := dec.Decode(benchDoc, &cfg); err != nil {
+			b.Fatal(err)
+		}
+	}
+}
+
+func BenchmarkParseLong(b *testing.B) {
+	b.ReportAllocs()
+	b.SetBytes(int64(len(longDoc)))
+	for b.Loop() {
+		if _, err := Parse(longDoc); err != nil {
+			b.Fatal(err)
+		}
+	}
+}