diff --git a/async-openai/src/types/types.rs b/async-openai/src/types/types.rs
index b99c4134..03672a00 100644
--- a/async-openai/src/types/types.rs
+++ b/async-openai/src/types/types.rs
@@ -869,12 +869,17 @@ pub type ChatCompletionResponseStream =
     Pin<Box<dyn Stream<Item = Result<CreateChatCompletionStreamResponse, OpenAIError>> + Send>>;
 
 // For reason (not documented by OpenAI) the response from stream is different
+#[derive(Debug, Deserialize, Serialize, Clone, PartialEq)]
+pub struct FunctionCallStream {
+    pub name: Option<String>,
+    pub arguments: Option<String>,
+}
 
 #[derive(Debug, Deserialize, Serialize, Clone, PartialEq)]
 pub struct ChatCompletionStreamResponseDelta {
     pub role: Option<Role>,
     pub content: Option<String>,
-    pub function_call: Option<FunctionCall>,
+    pub function_call: Option<FunctionCallStream>,
 }
 
 #[derive(Debug, Deserialize, Serialize, Clone, PartialEq)]
diff --git a/examples/function-call-stream/src/main.rs b/examples/function-call-stream/src/main.rs
index ffe8cce6..0570e31e 100644
--- a/examples/function-call-stream/src/main.rs
+++ b/examples/function-call-stream/src/main.rs
@@ -12,6 +12,7 @@ use async_openai::{
 
 use futures::StreamExt;
 use serde_json::json;
+use async_openai::config::OpenAIConfig;
 
 #[tokio::main]
 async fn main() -> Result<(), Box<dyn Error>> {
@@ -42,71 +43,95 @@ async fn main() -> Result<(), Box<dyn Error>> {
         .function_call("auto")
         .build()?;
 
-    // the first response from GPT is just the json response containing the function that was called
-    // and the model-generated arguments for that function (don't stream this)
-    let response = client
-        .chat()
-        .create(request)
-        .await?
-        .choices
-        .get(0)
-        .unwrap()
-        .message
-        .clone();
-
-    if let Some(function_call) = response.function_call {
-        let mut available_functions: HashMap<&str, fn(&str, &str) -> serde_json::Value> =
-            HashMap::new();
-        available_functions.insert("get_current_weather", get_current_weather);
-
-        let function_name = function_call.name;
-        let function_args: serde_json::Value = function_call.arguments.parse().unwrap();
-
-        let location = function_args["location"].as_str().unwrap();
-        let unit = "fahrenheit"; // why doesn't the model return a unit argument?
-        let function = available_functions.get(function_name.as_str()).unwrap();
-        let function_response = function(location, unit); // call the function
-
-        let message = vec![
-            ChatCompletionRequestMessageArgs::default()
-                .role(Role::User)
-                .content("What's the weather like in Boston?")
-                .build()?,
-            ChatCompletionRequestMessageArgs::default()
-                .role(Role::Function)
-                .content(function_response.to_string())
-                .name(function_name)
-                .build()?,
-        ];
-
-        let request = CreateChatCompletionRequestArgs::default()
-            .max_tokens(512u16)
-            .model("gpt-3.5-turbo-0613")
-            .messages(message)
-            .build()?;
-
-        // Now stream received response from model, which essentially formats the function response
-        let mut stream = client.chat().create_stream(request).await?;
-
-        let mut lock = stdout().lock();
-        while let Some(result) = stream.next().await {
-            match result {
-                Ok(response) => {
-                    response.choices.iter().for_each(|chat_choice| {
-                        if let Some(ref content) = chat_choice.delta.content {
-                            write!(lock, "{}", content).unwrap();
+    let mut stream = client.chat().create_stream(request).await?;
+
+    let mut fn_name = String::new();
+    let mut fn_args = String::new();
+
+    let mut lock = stdout().lock();
+    while let Some(result) = stream.next().await {
+        match result {
+            Ok(response) => {
+                for chat_choice in response.choices {
+                    if let Some(fn_call) = &chat_choice.delta.function_call {
+                        writeln!(lock, "function_call: {:?}", fn_call).unwrap();
+                        if let Some(name) = &fn_call.name {
+                            fn_name = name.clone();
                         }
-                    });
-                }
-                Err(err) => {
-                    writeln!(lock, "error: {err}").unwrap();
+                        if let Some(args) = &fn_call.arguments {
+                            fn_args.push_str(args);
+                        }
+                    }
+                    if let Some(finish_reason) = &chat_choice.finish_reason {
+                        if finish_reason == "function_call" {
+                            call_fn(&client, &fn_name, &fn_args).await?;
+                        }
+                    } else if let Some(content) = &chat_choice.delta.content {
+                        write!(lock, "{}", content).unwrap();
+                    }
                 }
             }
-            stdout().flush()?;
+            Err(err) => {
+                writeln!(lock, "error: {err}").unwrap();
+            }
         }
-        println!("{}", "\n");
+        stdout().flush()?;
     }
 
+
+    Ok(())
+}
+
+async fn call_fn(client: &Client<OpenAIConfig>, name: &str, args: &str) -> Result<(), Box<dyn Error>> {
+    let mut available_functions: HashMap<&str, fn(&str, &str) -> serde_json::Value> =
+        HashMap::new();
+    available_functions.insert("get_current_weather", get_current_weather);
+
+    let function_args: serde_json::Value = args.parse().unwrap();
+
+    let location = function_args["location"].as_str().unwrap();
+    let unit = function_args["unit"].as_str().unwrap_or("fahrenheit");
+    let function = available_functions.get(name).unwrap();
+    let function_response = function(location, unit); // call the function
+
+    let message = vec![
+        ChatCompletionRequestMessageArgs::default()
+            .role(Role::User)
+            .content("What's the weather like in Boston?")
+            .build()?,
+        ChatCompletionRequestMessageArgs::default()
+            .role(Role::Function)
+            .content(function_response.to_string())
+            .name(name.clone())
+            .build()?,
+    ];
+
+    let request = CreateChatCompletionRequestArgs::default()
+        .max_tokens(512u16)
+        .model("gpt-3.5-turbo-0613")
+        .messages(message)
+        .build()?;
+
+    // Now stream received response from model, which essentially formats the function response
+    let mut stream = client.chat().create_stream(request).await?;
+
+    let mut lock = stdout().lock();
+    while let Some(result) = stream.next().await {
+        match result {
+            Ok(response) => {
+                response.choices.iter().for_each(|chat_choice| {
+                    if let Some(ref content) = chat_choice.delta.content {
+                        write!(lock, "{}", content).unwrap();
+                    }
+                });
+            }
+            Err(err) => {
+                writeln!(lock, "error: {err}").unwrap();
+            }
+        }
+        stdout().flush()?;
+    }
+    println!("{}", "\n");
     Ok(())
 }