diff --git a/Cargo.toml b/Cargo.toml
index 39fa3f2..9515634 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -32,6 +32,7 @@ tracing = { version = "0.1", features = ["log"] }
 [dev-dependencies]
 assert_matches = "1.5"
 clap = { version = "4", features = ["derive"] }
+criterion = { version = "0.5", features = ["async_tokio"] }
 maplit = "1"
 pretty_env_logger = "0.5"
 rustyline = "11"
@@ -39,6 +40,7 @@ serde_json = "1"
 tokio = { version = "1", features = ["full"] }
 tracing-test = { version = "0.2", features = ["no-env-filter"] }
 tarantool-test-container = { path = "tarantool-test-container" }
+rusty_tarantool = "*"
 
 [[example]]
 name = "cli_client"
@@ -52,6 +54,14 @@ name = "transactions"
 [[example]]
 name = "call"
 
+[[bench]]
+name = "bench"
+harness = false
+
+[[bench]]
+name = "compare"
+harness = false
+
 [workspace]
 members = [
     "tarantool-test-container",
diff --git a/benches/bench.rs b/benches/bench.rs
new file mode 100644
index 0000000..931b62a
--- /dev/null
+++ b/benches/bench.rs
@@ -0,0 +1,67 @@
+use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion};
+use futures::{future::try_join_all, TryFutureExt};
+use tarantool_rs::{Connection, ExecutorExt};
+
+type TarantoolTestContainer = tarantool_test_container::TarantoolTestContainer<
+    tarantool_test_container::TarantoolDefaultArgs,
+>;
+
+pub fn bench_tarantool_rs(c: &mut Criterion) {
+    let mut group = c.benchmark_group("tarantool_rs");
+
+    // Preparations
+    let tokio_rt = tokio::runtime::Builder::new_multi_thread()
+        .enable_all()
+        .build()
+        .expect("Tokio multithread runtime built");
+    let container = TarantoolTestContainer::default();
+    let conn = tokio_rt
+        .block_on(async {
+            Connection::builder()
+                .build(format!("127.0.0.1:{}", container.connect_port()))
+                .await
+        })
+        .expect("Connection created successfully");
+
+    // Bench logic
+    // NOTE: on my PC converting to join add slight overhead (1-2 microseconds for 1 future input)
+    // NOTE: on my PC 50 input load tarantool to 50% on single core
+    for parallel in [1, 2, 5, 10, 50].into_iter() {
+        group.bench_with_input(BenchmarkId::new("ping", parallel), &parallel, |b, p| {
+            b.to_async(&tokio_rt).iter(|| async {
+                let make_fut = |_| conn.ping();
+                let futures = try_join_all((0..*p).map(make_fut));
+                futures.await.expect("Successful bench");
+            })
+        });
+        group.bench_with_input(BenchmarkId::new("call", parallel), &parallel, |b, p| {
+            b.to_async(&tokio_rt).iter(|| async {
+                let make_fut = |_| conn.call("gcinfo", ());
+                let futures = try_join_all((0..*p).map(make_fut));
+                futures.await.expect("Successful bench");
+            })
+        });
+        group.bench_with_input(
+            BenchmarkId::new("eval, pass and decode echo response", parallel),
+            &parallel,
+            |b, p| {
+                b.to_async(&tokio_rt).iter(|| async {
+                    let make_fut = |_| {
+                        conn.eval("return ...", (1, "two", true))
+                            .and_then(|resp| async {
+                                resp.decode_full::<(i64, String, bool)>()
+                                    .map_err(Into::into)
+                            })
+                    };
+                    let futures = try_join_all((0..*p).map(make_fut));
+                    futures.await.expect("Successful bench");
+                })
+            },
+        );
+    }
+
+    group.finish();
+}
+
+criterion_group!(benches, bench_tarantool_rs,);
+criterion_main!(benches);
diff --git a/benches/compare.rs b/benches/compare.rs
new file mode 100644
index 0000000..167a99d
--- /dev/null
+++ b/benches/compare.rs
@@ -0,0 +1,98 @@
+use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion};
+use futures::{future::try_join_all, TryFutureExt};
+use tarantool_rs::{Connection, ExecutorExt};
+
+type TarantoolTestContainer = tarantool_test_container::TarantoolTestContainer<
+    tarantool_test_container::TarantoolDefaultArgs,
+>;
+
+pub fn compare_tarantool_rs_and_rusty_tarantool(c: &mut Criterion) {
+    let mut group = c.benchmark_group("compare_with_rusty_tarantool");
+
+    // Preparations
+    let tokio_rt = tokio::runtime::Builder::new_multi_thread()
+        .enable_all()
+        .build()
+        .expect("Tokio multithread runtime built");
+    let container = TarantoolTestContainer::default();
+    let tt_addr = || format!("127.0.0.1:{}", container.connect_port());
+    let conn = tokio_rt
+        .block_on(async { Connection::builder().build(tt_addr()).await })
+        .expect("Connection created successfully");
+    let conn_rusty = tokio_rt
+        .block_on(async {
+            let conn =
+                rusty_tarantool::tarantool::ClientConfig::new(tt_addr(), "guest", "").build();
+            conn.ping().await?;
+            Result::<_, anyhow::Error>::Ok(conn)
+        })
+        .expect("Connection (RustyTarantool) created successfully");
+
+    // Bench logic
+    // NOTE: on my PC converting to join add slight overhead (1-2 microseconds for 1 future input)
+    // NOTE: on my PC 50 input load tarantool to 50% on single core
+    for parallel in [1, 2, 5, 10, 50].into_iter() {
+        group.bench_with_input(
+            BenchmarkId::new("tarantool_rs ping", parallel),
+            &parallel,
+            |b, p| {
+                b.to_async(&tokio_rt).iter(|| async {
+                    let make_fut = |_| conn.ping();
+                    let futures = try_join_all((0..*p).map(make_fut));
+                    futures.await.expect("Successful bench");
+                })
+            },
+        );
+        group.bench_with_input(
+            BenchmarkId::new("tarantool_rs eval", parallel),
+            &parallel,
+            |b, p| {
+                b.to_async(&tokio_rt).iter(|| async {
+                    let make_fut = |_| {
+                        conn.eval("return ...", (1, "two", true))
+                            .and_then(|resp| async {
+                                resp.decode_full::<(i64, String, bool)>()
+                                    .map_err(Into::into)
+                            })
+                    };
+                    let futures = try_join_all((0..*p).map(make_fut));
+                    futures.await.expect("Successful bench");
+                })
+            },
+        );
+
+        group.bench_with_input(
+            BenchmarkId::new("RustyTarantool ping", parallel),
+            &parallel,
+            |b, p| {
+                b.to_async(&tokio_rt).iter(|| async {
+                    let make_fut = |_| conn_rusty.ping();
+                    let futures = try_join_all((0..*p).map(make_fut));
+                    futures.await.expect("Successful bench");
+                })
+            },
+        );
+        group.bench_with_input(
+            BenchmarkId::new("RustyTarantool eval", parallel),
+            &parallel,
+            |b, p| {
+                b.to_async(&tokio_rt).iter(|| async {
+                    let make_fut = |_| {
+                        conn_rusty
+                            .eval("return ...", &(1, "two", true))
+                            .and_then(|resp| async {
+                                resp.decode::<(i64, String, bool)>().map_err(Into::into)
+                            })
+                    };
+                    let futures = try_join_all((0..*p).map(make_fut));
+                    futures.await.expect("Successful bench");
+                })
+            },
+        );
+    }
+
+    group.finish();
+}
+
+criterion_group!(benches, compare_tarantool_rs_and_rusty_tarantool);
+criterion_main!(benches);