diff --git a/native/spark-expr/Cargo.toml b/native/spark-expr/Cargo.toml
index 6ce2835c60a..7d12e9afa7f 100644
--- a/native/spark-expr/Cargo.toml
+++ b/native/spark-expr/Cargo.toml
@@ -407,6 +407,10 @@ harness = false
 name = "dayofweek_weekday"
 harness = false
 
+[[bench]]
+name = "aggregate_stats"
+harness = false
+
 [[bench]]
 name = "nondeterministic"
 harness = false
diff --git a/native/spark-expr/benches/aggregate_stats.rs b/native/spark-expr/benches/aggregate_stats.rs
new file mode 100644
index 00000000000..edc581d3fe3
--- /dev/null
+++ b/native/spark-expr/benches/aggregate_stats.rs
@@ -0,0 +1,287 @@
+// 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.
+
+//! Benchmarks for Comet-owned `agg_funcs` accumulators not covered by
+//! `aggregate.rs` (which covers `avg_decimal` / `sum_decimal` / `sum_int`):
+//! the Welford-path statistical aggregates (`variance`, `covariance`,
+//! `correlation`), non-decimal `avg`, exact and approximate percentile, and
+//! HyperLogLog++ `approx_count_distinct`. Only Comet's own accumulators are
+//! benched here. `stddev` is intentionally omitted: `StddevAccumulator` wraps
+//! `VarianceAccumulator`, so its per-row update loop is identical to `variance`
+//! (it only adds a final `sqrt`).
+
+use arrow::array::{ArrayRef, Float64Builder, RecordBatch, StringBuilder};
+use arrow::datatypes::SchemaRef;
+use arrow::datatypes::{DataType, Field, Schema};
+use criterion::{criterion_group, criterion_main, Criterion};
+use datafusion::datasource::memory::MemorySourceConfig;
+use datafusion::datasource::source::DataSourceExec;
+use datafusion::execution::TaskContext;
+use datafusion::logical_expr::AggregateUDF;
+use datafusion::physical_expr::aggregate::AggregateExprBuilder;
+use datafusion::physical_expr::expressions::{lit, Column, StatsType};
+use datafusion::physical_expr::PhysicalExpr;
+use datafusion::physical_plan::aggregates::{AggregateExec, AggregateMode, PhysicalGroupBy};
+use datafusion::physical_plan::ExecutionPlan;
+use datafusion_comet_spark_expr::{
+    ApproxPercentile, Avg, Correlation, Covariance, HllPlusPlus, SparkPercentile, Variance,
+};
+use futures::StreamExt;
+use std::hint::black_box;
+use std::sync::Arc;
+use std::time::Duration;
+use tokio::runtime::Runtime;
+
+fn criterion_benchmark(c: &mut Criterion) {
+    let num_rows = 8192;
+    let batch = create_float_record_batch(num_rows);
+    let mut batches = Vec::new();
+    for _ in 0..10 {
+        batches.push(batch.clone());
+    }
+    let partitions = &[batches];
+
+    let c0: Arc<dyn PhysicalExpr> = Arc::new(Column::new("c0", 0));
+    let c1: Arc<dyn PhysicalExpr> = Arc::new(Column::new("c1", 1));
+    let c2: Arc<dyn PhysicalExpr> = Arc::new(Column::new("c2", 2));
+
+    let rt = Runtime::new().unwrap();
+
+    // Single-input accumulators: variance (population + sample) and avg.
+    let mut group = c.benchmark_group("stats_agg_single");
+
+    let single_cases: Vec<(&str, Arc<AggregateUDF>)> = vec![
+        ("variance_samp", comet_variance(StatsType::Sample)),
+        ("variance_pop", comet_variance(StatsType::Population)),
+        (
+            "avg",
+            Arc::new(AggregateUDF::new_from_impl(Avg::new(
+                "avg",
+                DataType::Float64,
+            ))),
+        ),
+    ];
+
+    for (name, udf) in single_cases {
+        group.bench_function(name, |b| {
+            b.to_async(&rt).iter(|| {
+                black_box(agg_test(
+                    partitions,
+                    c0.clone(),
+                    vec![c1.clone()],
+                    udf.clone(),
+                    name,
+                ))
+            })
+        });
+    }
+    group.finish();
+
+    // Two-input accumulators: covariance (population + sample) / correlation.
+    let mut group = c.benchmark_group("stats_agg_pair");
+
+    let pair_cases: Vec<(&str, Arc<AggregateUDF>)> = vec![
+        ("covariance_samp", comet_covariance(StatsType::Sample)),
+        ("covariance_pop", comet_covariance(StatsType::Population)),
+        ("correlation", comet_correlation()),
+    ];
+
+    for (name, udf) in pair_cases {
+        group.bench_function(name, |b| {
+            b.to_async(&rt).iter(|| {
+                black_box(agg_test(
+                    partitions,
+                    c0.clone(),
+                    vec![c1.clone(), c2.clone()],
+                    udf.clone(),
+                    name,
+                ))
+            })
+        });
+    }
+    group.finish();
+
+    // Percentile and approximate distinct count.
+    let mut group = c.benchmark_group("stats_agg_percentile");
+
+    group.bench_function("approx_percentile", |b| {
+        let udf = Arc::new(AggregateUDF::new_from_impl(ApproxPercentile::new(
+            vec![0.5],
+            10000,
+            DataType::Float64,
+            false,
+        )));
+        b.to_async(&rt).iter(|| {
+            black_box(agg_test(
+                partitions,
+                c0.clone(),
+                vec![c1.clone()],
+                udf.clone(),
+                "approx_percentile",
+            ))
+        })
+    });
+
+    group.bench_function("percentile", |b| {
+        let udf = Arc::new(AggregateUDF::new_from_impl(
+            SparkPercentile::try_new(0.5).unwrap(),
+        ));
+        // Mirror the planner: exact percentile takes the value column plus the
+        // percentage literal as inputs.
+        let inputs: Vec<Arc<dyn PhysicalExpr>> = vec![c1.clone(), lit(0.5f64)];
+        b.to_async(&rt).iter(|| {
+            black_box(agg_test(
+                partitions,
+                c0.clone(),
+                inputs.clone(),
+                udf.clone(),
+                "percentile",
+            ))
+        })
+    });
+
+    for p in [10i32, 14] {
+        group.bench_function(format!("approx_count_distinct_p{p}"), |b| {
+            let udf = Arc::new(AggregateUDF::new_from_impl(HllPlusPlus::new(p)));
+            b.to_async(&rt).iter(|| {
+                black_box(agg_test(
+                    partitions,
+                    c0.clone(),
+                    vec![c1.clone()],
+                    udf.clone(),
+                    "approx_count_distinct",
+                ))
+            })
+        });
+    }
+    group.finish();
+}
+
+fn comet_variance(stats_type: StatsType) -> Arc<AggregateUDF> {
+    Arc::new(AggregateUDF::new_from_impl(Variance::new(
+        "variance",
+        DataType::Float64,
+        stats_type,
+        false,
+    )))
+}
+
+fn comet_covariance(stats_type: StatsType) -> Arc<AggregateUDF> {
+    Arc::new(AggregateUDF::new_from_impl(Covariance::new(
+        "covariance",
+        DataType::Float64,
+        stats_type,
+        false,
+    )))
+}
+
+fn comet_correlation() -> Arc<AggregateUDF> {
+    Arc::new(AggregateUDF::new_from_impl(Correlation::new(
+        "correlation",
+        DataType::Float64,
+        false,
+    )))
+}
+
+async fn agg_test(
+    partitions: &[Vec<RecordBatch>],
+    group_col: Arc<dyn PhysicalExpr>,
+    input_exprs: Vec<Arc<dyn PhysicalExpr>>,
+    aggregate_udf: Arc<AggregateUDF>,
+    alias: &str,
+) {
+    let schema = &partitions[0][0].schema();
+    let scan: Arc<dyn ExecutionPlan> = Arc::new(DataSourceExec::new(Arc::new(
+        MemorySourceConfig::try_new(partitions, Arc::clone(schema), None).unwrap(),
+    )));
+    let aggregate = create_aggregate(scan, group_col, input_exprs, schema, aggregate_udf, alias);
+    let mut stream = aggregate
+        .execute(0, Arc::new(TaskContext::default()))
+        .unwrap();
+    while let Some(batch) = stream.next().await {
+        let _batch = batch.unwrap();
+    }
+}
+
+fn create_aggregate(
+    scan: Arc<dyn ExecutionPlan>,
+    group_col: Arc<dyn PhysicalExpr>,
+    input_exprs: Vec<Arc<dyn PhysicalExpr>>,
+    schema: &SchemaRef,
+    aggregate_udf: Arc<AggregateUDF>,
+    alias: &str,
+) -> Arc<AggregateExec> {
+    let aggr_expr = AggregateExprBuilder::new(aggregate_udf, input_exprs)
+        .schema(schema.clone())
+        .alias(alias)
+        .with_ignore_nulls(false)
+        .with_distinct(false)
+        .build()
+        .unwrap();
+
+    Arc::new(
+        AggregateExec::try_new(
+            AggregateMode::Partial,
+            PhysicalGroupBy::new_single(vec![(group_col, "c0".to_string())]),
+            vec![aggr_expr.into()],
+            vec![None], // no filter expressions
+            scan,
+            Arc::clone(schema),
+        )
+        .unwrap(),
+    )
+}
+
+fn create_float_record_batch(num_rows: usize) -> RecordBatch {
+    let mut c1_builder = Float64Builder::with_capacity(num_rows);
+    let mut c2_builder = Float64Builder::with_capacity(num_rows);
+    let mut string_builder = StringBuilder::with_capacity(num_rows, num_rows * 8);
+    for i in 0..num_rows {
+        // Spread-out deterministic values: representative (non-constant) input
+        // that also avoids the degenerate zero-variance / undefined-correlation
+        // cases.
+        c1_builder.append_value(i as f64 * 1.5 + 0.25);
+        c2_builder.append_value((num_rows - i) as f64 * 0.75 - 0.5);
+        string_builder.append_value(format!("group_{}", i % 1024));
+    }
+
+    let fields = vec![
+        Field::new("c0", DataType::Utf8, false),
+        Field::new("c1", DataType::Float64, false),
+        Field::new("c2", DataType::Float64, false),
+    ];
+    let columns: Vec<ArrayRef> = vec![
+        Arc::new(string_builder.finish()),
+        Arc::new(c1_builder.finish()),
+        Arc::new(c2_builder.finish()),
+    ];
+
+    RecordBatch::try_new(Arc::new(Schema::new(fields)), columns).unwrap()
+}
+
+fn config() -> Criterion {
+    Criterion::default()
+        .measurement_time(Duration::from_millis(500))
+        .warm_up_time(Duration::from_millis(500))
+}
+
+criterion_group! {
+    name = benches;
+    config = config();
+    targets = criterion_benchmark
+}
+criterion_main!(benches);