diff --git a/.gitignore b/.gitignore
index c8e84af..b07c3d3 100644
--- a/.gitignore
+++ b/.gitignore
@@ -7,3 +7,4 @@ src/*.so
 src/*.dll
 scratch.R
 docs
+inst/doc
diff --git a/DESCRIPTION b/DESCRIPTION
index 151c5f5..5cd4a64 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -14,7 +14,12 @@ LazyData: true
 LinkingTo: 
     cpp11
 Suggests: 
+    bench,
+    ggplot2,
+    knitr,
+    rmarkdown,
     testthat (>= 3.0.0)
 Config/testthat/edition: 3
 RoxygenNote: 7.2.3
 URL: https://parmsam.github.io/lzstring-r/
+VignetteBuilder: knitr
diff --git a/vignettes/.gitignore b/vignettes/.gitignore
new file mode 100644
index 0000000..097b241
--- /dev/null
+++ b/vignettes/.gitignore
@@ -0,0 +1,2 @@
+*.html
+*.R
diff --git a/vignettes/performance.Rmd b/vignettes/performance.Rmd
new file mode 100644
index 0000000..6275c1c
--- /dev/null
+++ b/vignettes/performance.Rmd
@@ -0,0 +1,66 @@
+---
+title: "Benchmarking performance"
+output: rmarkdown::html_vignette
+vignette: >
+  %\VignetteIndexEntry{performance}
+  %\VignetteEncoding{UTF-8}
+  %\VignetteEngine{knitr::rmarkdown}
+editor_options:
+  chunk_output_type: console
+---
+
+```{r, include = FALSE}
+knitr::opts_chunk$set(
+  collapse = TRUE,
+  comment = "#>"
+)
+```
+
+```{r setup}
+library(lzstring)
+library(bench)
+library(ggplot2)
+```
+
+Let's benchmark use of lzstring for compression to base 64 and decompression from base 64. We will generate random strings of different lengths and measure the time taken for compression and decompression.
+
+```{r benchmark}
+# Function to generate a random string of specified length
+generate_random_string <- function(length) {
+  paste(sample(c(LETTERS, 0:9), size = length, replace = TRUE), collapse = "")
+}
+
+# Prepare a data frame to store benchmark results
+benchmark_results <- data.frame(length_str = numeric(), comp_time = numeric(), decomp_time = numeric())
+
+# Run benchmark for each string length
+for (i in 1:100) {
+  big_str <- generate_random_string(i * 100)
+
+  comp <- mark(
+    compressToBase64(big_str)
+  )
+  decomp <- mark(
+    decompressFromBase64(comp$result[[1]])
+  )
+  decomp_time_tmp <- as.numeric(decomp$median)
+  comp_time_tmp <- as.numeric(comp$median)
+  # Store the median time and string length
+  benchmark_results <- rbind(benchmark_results, data.frame(length_str = nchar(big_str), comp_time = comp_time_tmp, decomp_time = decomp_time_tmp))
+}
+# Convert time to milliseconds
+benchmark_results$comp_time <- benchmark_results$comp_time * 1000
+benchmark_results$decomp_time <- benchmark_results$decomp_time * 1000
+```
+
+Now let's plot the results for compression and decompression times of all those strings we created.
+
+```{r plot-results, fig.width = 7, fig.asp = 0.6}
+ggplot(benchmark_results) +
+  geom_line(aes(x = length_str, y = comp_time, col = "Compression"), ) +
+  geom_line(aes(x = length_str, y = decomp_time, col = "Decompression")) +
+  labs(x = "String length (characters)",
+       y = "Time (milliseconds)") +
+  theme_minimal() +
+  theme(legend.position = "bottom")
+```