diff --git a/README.md b/README.md
index 14d8383..ff28d59 100644
--- a/README.md
+++ b/README.md
@@ -35,6 +35,7 @@ Feel free to run the program multiple times to get the best result!
 - `-o`, `--output` : Specify an output file
 - `-h`, `--help` : Print help
 - `-V`, `--version` : Print version
+- `-t`, `--token-usage` : Print token usage
 
 ## 🛠️ Building
 
diff --git a/src/api/chat_completions.rs b/src/api/chat_completions.rs
index 9306b34..faa7e43 100644
--- a/src/api/chat_completions.rs
+++ b/src/api/chat_completions.rs
@@ -1,8 +1,10 @@
-use super::json_models::chat_completion::ChatCompletionResponse;
+use super::json_models::chat_completion::{ChatCompletionResponse};
 use super::Instance;
 use crate::clue::ClueCollection;
 use serde_json::json;
 
+
+
 const SYSTEM_PROMPT: &str = r#"
 You are the spymaster in Codenames.
 I will give you a list of [agent word], followed by a list of [avoid word].
@@ -38,12 +40,20 @@ impl Instance {
             .await
             .map_err(|_| "Failed to fetch clue collection from API server")?;
 
-        // Deserialize the response
-        let clue_strings = response
+        let parsed_response = response
             .json::<ChatCompletionResponse>()
             .await
-            .map_err(|_| "Failed to parse clues from API server")?
-            .choices[0]
+            .map_err(|_| "Failed to parse clues from API server")?;
+
+        // Extract usage information from the parsed response
+        let token_usage = parsed_response.usage;
+
+
+        // Extract clue strings from the parsed response
+        let clue_strings = parsed_response
+            .choices
+            .get(0)
+            .ok_or("No choices returned from API")?
             .message
             .content
             .lines()
@@ -51,7 +61,7 @@ impl Instance {
             .collect::<Vec<String>>();
 
         // Build clues
-        let clue_collection = ClueCollection::new(clue_strings);
+        let clue_collection = ClueCollection::new(clue_strings, token_usage);
 
         Ok(clue_collection)
     }
diff --git a/src/api/json_models/chat_completion.rs b/src/api/json_models/chat_completion.rs
index 47fd4b2..08e1469 100644
--- a/src/api/json_models/chat_completion.rs
+++ b/src/api/json_models/chat_completion.rs
@@ -10,7 +10,15 @@ pub struct Choice {
     pub message: Message,
 }
 
+#[derive(Deserialize)]
+pub struct Usage {
+    pub prompt_tokens: usize,
+    pub completion_tokens: usize,
+    pub total_tokens: usize,
+}
+
 #[derive(Deserialize)]
 pub struct ChatCompletionResponse {
     pub choices: Vec<Choice>,
+    pub usage: Usage,
 }
diff --git a/src/api/json_models/mod.rs b/src/api/json_models/mod.rs
index 7381405..c19183e 100644
--- a/src/api/json_models/mod.rs
+++ b/src/api/json_models/mod.rs
@@ -1,2 +1,2 @@
-pub mod chat_completion;
 pub mod language_model;
+pub mod chat_completion;
diff --git a/src/api/mod.rs b/src/api/mod.rs
index b95b552..8b9fa6f 100644
--- a/src/api/mod.rs
+++ b/src/api/mod.rs
@@ -1,5 +1,5 @@
-mod chat_completions;
-mod json_models;
+pub mod chat_completions;
+pub mod json_models;
 mod language_models;
 
 use dotenv::dotenv;
diff --git a/src/clue.rs b/src/clue.rs
index 3f53ce9..4736355 100644
--- a/src/clue.rs
+++ b/src/clue.rs
@@ -1,7 +1,7 @@
 use comfy_table::modifiers::UTF8_ROUND_CORNERS;
 use comfy_table::presets::UTF8_FULL;
 use comfy_table::{Attribute, Cell, CellAlignment, ContentArrangement, Table};
-
+use crate::api::json_models::chat_completion::Usage;
 struct Clue {
     clue_word: String,
     count: usize,
@@ -10,6 +10,7 @@ struct Clue {
 
 pub struct ClueCollection {
     clues: Vec<Clue>,
+    pub usage: Usage,
 }
 
 impl Clue {
@@ -17,6 +18,8 @@ impl Clue {
     pub fn new(clue_line: &str) -> Option<Self> {
         let chunks: Vec<&str> = clue_line.split(", ").collect();
 
+
+
         // Discard empty lines as well as clues with only one word linked
         if chunks.len() < 4 {
             return None;
@@ -44,13 +47,13 @@ impl Clue {
 
 impl ClueCollection {
     /// Create an instance of `ClueCollection` from `Vec<String>`, which contains lines of clue response from the API
-    pub fn new(clue_strings: Vec<String>) -> Self {
+    pub fn new(clue_strings: Vec<String>, usage: Usage) -> Self {
         let mut clues: Vec<Clue> = clue_strings.iter().filter_map(|s| Clue::new(s)).collect();
 
         // Sort the clues by the number of words they link together
         clues.sort_by(|a, b| b.count.cmp(&a.count));
 
-        Self { clues }
+        Self { clues, usage }
     }
 
     pub fn is_empty(&self) -> bool {
diff --git a/src/lib.rs b/src/lib.rs
index d66cda0..14e64f0 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -21,6 +21,10 @@ pub struct Args {
     #[arg(short, long, value_name = "FILE")]
     pub output: Option<PathBuf>,
 
+    /// Print all token usage information
+    #[arg(short, long = "token-usage")]
+    pub token: bool,
+
     /// File containing words to link together - the words from your team
     #[arg(required_unless_present = "get")]
     pub to_link: Option<PathBuf>,
diff --git a/src/main.rs b/src/main.rs
index fbdf4f0..983c8f3 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -31,6 +31,7 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
         .fetch_clue_collection(link_words, avoid_words)
         .await?;
 
+
     // Output
     if clue_collection.is_empty() {
         println!("The language model didn't return any useful clues. Maybe try again?");
@@ -41,5 +42,12 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
         clue_collection.display();
     }
 
+    // If -t is set, output the token usage information
+    if args.token {
+        // Write to stderr in the format: prompt_tokens, completion_tokens, total_tokens
+        eprintln!("\nTokens Usage\n----------------------\nPrompt Tokens: {}\nCompletion Tokens: {}\n----------------------\nTotal Tokens: {}",
+                  clue_collection.usage.prompt_tokens, clue_collection.usage.completion_tokens, clue_collection.usage.total_tokens);
+    }
+
     Ok(())
 }