diff --git a/src/iceberg/benchmark/CMakeLists.txt b/src/iceberg/benchmark/CMakeLists.txt
index 5d3c19d0f5..a34995ff1f 100644
--- a/src/iceberg/benchmark/CMakeLists.txt
+++ b/src/iceberg/benchmark/CMakeLists.txt
@@ -17,8 +17,13 @@
 
 # Google Benchmark is resolved in cmake_modules/IcebergThirdpartyToolchain.cmake.
 
-add_executable(benchmark_smoke benchmark_smoke.cc)
-target_link_libraries(benchmark_smoke
-                      PRIVATE "$<IF:$<TARGET_EXISTS:iceberg_static>,iceberg_static,iceberg_shared>"
-                              "$<IF:$<TARGET_EXISTS:benchmark::benchmark>,benchmark::benchmark,benchmark>"
-)
+function(add_iceberg_benchmark name)
+  add_executable(${name} ${name}.cc)
+  target_link_libraries(${name}
+                        PRIVATE "$<IF:$<TARGET_EXISTS:iceberg_static>,iceberg_static,iceberg_shared>"
+                                "$<IF:$<TARGET_EXISTS:benchmark::benchmark>,benchmark::benchmark,benchmark>"
+  )
+endfunction()
+
+add_iceberg_benchmark(benchmark_smoke)
+add_iceberg_benchmark(metrics_evaluator_benchmark)
diff --git a/src/iceberg/benchmark/metrics_evaluator_benchmark.cc b/src/iceberg/benchmark/metrics_evaluator_benchmark.cc
new file mode 100644
index 0000000000..0560c0deb1
--- /dev/null
+++ b/src/iceberg/benchmark/metrics_evaluator_benchmark.cc
@@ -0,0 +1,210 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#include <cstdint>
+#include <memory>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include <benchmark/benchmark.h>
+
+#include "iceberg/expression/expression.h"
+#include "iceberg/expression/expressions.h"
+#include "iceberg/expression/inclusive_metrics_evaluator.h"
+#include "iceberg/expression/literal.h"
+#include "iceberg/expression/predicate.h"
+#include "iceberg/expression/strict_metrics_evaluator.h"
+#include "iceberg/manifest/manifest_entry.h"
+#include "iceberg/result.h"
+#include "iceberg/schema.h"
+#include "iceberg/schema_field.h"
+#include "iceberg/type.h"
+
+namespace iceberg {
+namespace {
+
+constexpr int32_t kNumColumns = 16;
+constexpr int64_t kRecordCount = 10'000;
+constexpr int64_t kNullCount = 10;
+
+enum class ColumnKind : int32_t { kLong = 0, kInt, kDouble, kString, kNumKinds };
+
+ColumnKind KindOf(int32_t field_id) {
+  return static_cast<ColumnKind>((field_id - 1) %
+                                 static_cast<int32_t>(ColumnKind::kNumKinds));
+}
+
+std::string ColumnName(int32_t field_id) { return "c" + std::to_string(field_id); }
+
+Literal MakeLiteral(int32_t field_id, int64_t number, const char* str) {
+  switch (KindOf(field_id)) {
+    case ColumnKind::kLong:
+      return Literal::Long(number);
+    case ColumnKind::kInt:
+      return Literal::Int(static_cast<int32_t>(number));
+    case ColumnKind::kDouble:
+      return Literal::Double(static_cast<double>(number));
+    default:
+      return Literal::String(str);
+  }
+}
+
+/// Predicates compare against a value inside the bounds, so every predicate has
+/// to deserialize and compare them.
+Literal LowerBound(int32_t field_id) { return MakeLiteral(field_id, 100, "abc"); }
+Literal UpperBound(int32_t field_id) { return MakeLiteral(field_id, 200, "xyz"); }
+Literal ValueInRange(int32_t field_id) { return MakeLiteral(field_id, 150, "mno"); }
+
+std::shared_ptr<Schema> MakeSchema() {
+  std::vector<SchemaField> fields;
+  fields.reserve(kNumColumns);
+  for (int32_t id = 1; id <= kNumColumns; ++id) {
+    std::shared_ptr<Type> type;
+    switch (KindOf(id)) {
+      case ColumnKind::kLong:
+        type = int64();
+        break;
+      case ColumnKind::kInt:
+        type = int32();
+        break;
+      case ColumnKind::kDouble:
+        type = float64();
+        break;
+      default:
+        type = string();
+        break;
+    }
+    // Optional, so that IS NULL and NOT NULL are not folded away during binding.
+    fields.push_back(SchemaField::MakeOptional(id, ColumnName(id), std::move(type)));
+  }
+  return std::make_shared<Schema>(std::move(fields), /*schema_id=*/0);
+}
+
+DataFile MakeDataFile() {
+  DataFile file;
+  file.file_path = "s3://bucket/warehouse/db/table/data/00000-0-data.parquet";
+  file.file_format = FileFormatType::kParquet;
+  file.record_count = kRecordCount;
+  file.file_size_in_bytes = int64_t{64} * 1024 * 1024;
+  for (int32_t id = 1; id <= kNumColumns; ++id) {
+    file.value_counts[id] = kRecordCount;
+    file.null_value_counts[id] = kNullCount;
+    if (KindOf(id) == ColumnKind::kDouble) {
+      file.nan_value_counts[id] = 0;
+    }
+    file.lower_bounds[id] = LowerBound(id).Serialize().value();
+    file.upper_bounds[id] = UpperBound(id).Serialize().value();
+  }
+  return file;
+}
+
+std::shared_ptr<Expression> MakePredicate(int64_t index) {
+  const auto id = static_cast<int32_t>(index % kNumColumns) + 1;
+  auto name = ColumnName(id);
+  switch (index % 6) {
+    case 0:
+      return Expressions::LessThan(std::move(name), ValueInRange(id));
+    case 1:
+      return Expressions::GreaterThanOrEqual(std::move(name), ValueInRange(id));
+    case 2:
+      return Expressions::Equal(std::move(name), ValueInRange(id));
+    case 3:
+      return Expressions::NotEqual(std::move(name), ValueInRange(id));
+    case 4:
+      return Expressions::In(std::move(name),
+                             {LowerBound(id), ValueInRange(id), UpperBound(id)});
+    default:
+      return Expressions::NotNull(std::move(name));
+  }
+}
+
+std::shared_ptr<Expression> MakeFilter(int64_t num_predicates) {
+  if (num_predicates <= 0) {
+    return Expressions::AlwaysTrue();
+  }
+  std::shared_ptr<Expression> filter = MakePredicate(0);
+  for (int64_t i = 1; i < num_predicates; ++i) {
+    filter = Expressions::And(std::move(filter), MakePredicate(i));
+  }
+  return filter;
+}
+
+struct Inclusive {
+  static Result<std::unique_ptr<InclusiveMetricsEvaluator>> Make(
+      std::shared_ptr<Expression> filter, const std::shared_ptr<Schema>& schema) {
+    return InclusiveMetricsEvaluator::Make(std::move(filter), *schema);
+  }
+};
+
+struct Strict {
+  static Result<std::unique_ptr<StrictMetricsEvaluator>> Make(
+      std::shared_ptr<Expression> filter, const std::shared_ptr<Schema>& schema) {
+    return StrictMetricsEvaluator::Make(std::move(filter), schema);
+  }
+};
+
+/// Scan planning: ManifestReader reuses one evaluator for every data file.
+template <typename EvaluatorT>
+void RunEvaluate(benchmark::State& state) {
+  const auto schema = MakeSchema();
+  const auto file = MakeDataFile();
+  auto evaluator = EvaluatorT::Make(MakeFilter(state.range(0)), schema);
+  if (!evaluator.has_value()) {
+    state.SkipWithError(evaluator.error().message.c_str());
+    return;
+  }
+  // Check once outside the timed loop, so that an error path is never measured.
+  if (auto result = (*evaluator)->Evaluate(file); !result.has_value()) {
+    state.SkipWithError(result.error().message.c_str());
+    return;
+  }
+
+  for (auto _ : state) {
+    auto result = (*evaluator)->Evaluate(file);
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations());
+}
+
+void BM_InclusiveMetricsEvaluate(benchmark::State& state) {
+  RunEvaluate<Inclusive>(state);
+}
+
+void BM_StrictMetricsEvaluate(benchmark::State& state) { RunEvaluate<Strict>(state); }
+
+void PredicateCounts(benchmark::internal::Benchmark* bench) {
+  bench->ArgName("predicates")->RangeMultiplier(4)->Range(1, 64);
+}
+
+BENCHMARK(BM_InclusiveMetricsEvaluate)->Apply(PredicateCounts);
+BENCHMARK(BM_StrictMetricsEvaluate)->Apply(PredicateCounts);
+
+}  // namespace
+}  // namespace iceberg
+
+int main(int argc, char** argv) {
+  benchmark::Initialize(&argc, argv);
+  if (benchmark::ReportUnrecognizedArguments(argc, argv)) {
+    return 1;
+  }
+  benchmark::RunSpecifiedBenchmarks();
+  benchmark::Shutdown();
+  return 0;
+}