AttributeError: module 'openai' has no attribute 'ChatCompletion'

Getting this error just today, worked yesterday without issues.

1 Like

Came to this issue on Google Colab. The following works:

!pip3 install openai

from openai import OpenAI
from google.colab import userdata

client = OpenAI(
def llm_response(prompt):
    response =
    return response.choices[0].message.content

prompt = '''
    Classify the following review 
    as having either a positive or
    negative sentiment:

    The banana pudding was really tasty!

response = llm_response(prompt)

Perhaps when posting in this thread someone could spend thirty seconds of reading, install “openai classic”, and press the thanks button for the answer above…

pip install "openai<1.0.0"

Or alternately code for the new methods of the API library changes.

If OpenAI had given anyone a heads up instead of jumping from 1.0.0beta2 all the way to 1.1.1 internal and dumping wheels on those millions of developers, maybe a generous person could have written and put in a pull request for another cookbook notebook to be put up “how this all works without paying for a 3rd party code-conversion service, all the way from simple calls up to asyncio streaming multimodal multi-client with token-counting chat history client management


I had the same problem, while using skll library

The only solution was to install openai version 0.28.1

pip install openai==0.28.1

1 Like

thanks…this works for me…i m using ubuntu on wsl2 and vscode

The error you’re encountering indicates that ChatCompletion is not a subscriptable object, meaning you can’t use indexing ([]) directly on it. It seems like the response object is not a dictionary, but an instance of a ChatCompletion class.

When dealing with a class instance, you would typically access its attributes using dot notation. If you are using the OpenAI Python client, the attributes of the ChatCompletion object would be accessed accordingly. However, the output structure seems to suggest that it should be possible to subscript it if it were a dictionary.

Here is how I fixed this:

# Assuming 'response' is an instance of a ChatCompletion or similar class 

message_content = response.choices[0].message.content 


Using the period instead of the bracket worked.

1 Like

As marciobernardo1 has mentioned it above, try to use openai version 0.28.1 as the ChatCompletion attribute is still there and your code should be working as expected.

Hello guys.

I’m having a similar problem but I’m not using it that way, I’m using it with discord to create a bot but it’s not working

import os
import discord
import openai
from discord.ext import commands

intents = discord.Intents.default()
intents.message_content = True
intents.members = True
prefix = "!"
bot = commands.Bot(command_prefix=prefix, intents=intents)

openai.api_key = os.environ.get("key_chatgpt")

# OpenAI
class ChatBotCog(commands.Cog):
    def __init__(self, bot): = bot

    async def ask(self,ctx, *, question):
            conversation = [
                {"role": "system", "content": "MEU NOME E Role Aleatorio, E AGORA TAMBÉM SOU UM CHAT-BOT"},
                {"role": "user", "content": question}

            response = openai.Completion.create(

            await ctx.send(response.completion.choices[0].message.content)
        except Exception as error:
            await ctx.send(f"Ocorreu um erro: {str(error)}")
async def setup(bot):
    await bot.add_cog(ChatBotCog(bot))

this is the code and this is the problem

can you help me with this.

Follow your code , return an error:
APITimeoutError: Request timed out.
why is this

That has multiple bot-written problems. You are calling a completion endpoint with a chat model, and using the deprecated “engine” parameter. You are trying to extract a chat response from a completion reply. It is written for the old library.

It is offtopic from those that simply experience incompatibility when updating the openai library.

That’s because of the volume of requests hitting the API

WHERE is this file that everyone is talking about?

“openai” is a python library. There’s a quickstart you could have read, but let’s jump in.

If you have Python 3.8-3.11 installed on your system for compatibility, you can, at your command line or shell:

pip install --upgrade openai

to install the latest version of the openai python library (wheel) and its dependencies.

You then can run Python scripts, applications, or more advanced uses with the new v1.1 client object programming style introduced November 6:

from openai import OpenAI
client = OpenAI()

system = [{"role": "system", "content":
           "You are Jbot, a helpful AI assistant."}]
user = [{"role": "user", "content":
         "introduce Jbot"}]
chat = []

while not user[0]['content'] == "exit":
        response =
            messages=system + chat[-20:] + user,
            max_tokens=1000, top_p=0.9,
            stream = True,
    except Exception as err:
        print(f"Unexpected {err=}, {type(err)=}")
    reply = ""
    for part in response:
        word = part.choices[0].delta.content or ""
        reply += word
        print(word, end ="")
    chat += user + [{"role": "assistant", "content": reply}]
    user = [{"role": "user", "content": input("\nPrompt: ")}]

Don’t save your programs as, as that will break everything. Save this as

You must then finally obtain an API key from your (funded) API account, and set the key value as a OS environment variable OPENAI_API_KEY.


“openai” is a python library. There’s a quickstart you could have read, but let’s jump in.

LOL. The quickstart is super simple too. You explained it well though.

Seem to be getting the same problem, despite having an acceptable python version, and the latest version of openai.

Screenshot 2023-11-28 183205

Running the following code:
client = OpenAI(api_key=os.environ.get(“OPENAI_API_KEY”))

completion =
{“role”: “system”, “content”: “You are a poetic assistant, skilled in explaining complex programming concepts with creative flair.”},
{“role”: “user”, “content”: “Compose a poem that explains the concept of recursion in programming.”}


What you probably meant to write:

from openai import OpenAI

client = OpenAI()

completion =
            "role": "system",
            "content": "You are a poetic assistant, skilled in explaining complex programming concepts with creative flair.",
            "role": "user",
            "content": "Compose a poem that explains the concept of recursion in programming.",


The python library will use the environment OPENAI_API_KEY if it is set.

# ottieni tag title da gpt
# Modulo per l'interazione con l'API di OpenAI

from dotenv import load_dotenv
import os
load_dotenv()  # Carica le variabili d'ambiente dal file .env
#openai.api_key = os.getenv("openai.api_key")
from openai import OpenAI
client = OpenAI(api_key=os.getenv("openai.api_key"))

# setup richiesta gtp
# nuovo dataframe
merged_df8b = merged_df8.copy()

# creo dizionario delle risposte di gpt per evitare chiamate doppie
gpt_responses = {}
# Inizializza un contatore per il numero totale di token
total_tokens_used = 0
# Inizializza un contatore per le chiamate API effettive
actual_api_calls = 0

def improve_tag_title(row, gpt_responses):
    global total_tokens_used  , actual_api_calls      
    top_5_queries = row['top_5_queries']

    # Ottieni il tag title e l'URL dalla riga corrente
    tag_title = row["tag title"]
    page_url = row["page"]

    # Controllo se tag_title è presente e non vuoto
    if not tag_title:
        return None

    # Controllo per saltare la chiamata GPT quando comparison è 'greater' o check è 'Ok'
    #if row["comparison"] == "greater" or "Missing" not in row["check"]:
    #    #return row["tag title"]
    #    return None

    # Verifica se l'URL è già stato processato
    if page_url in gpt_responses:
        return gpt_responses[page_url]  # Usa il tag title migliorato se disponibile

    prompt = (      
        f"Sto cercando di migliorare il tag title per una pagina web. Il tag title attuale è: {tag_title}.\n"
        f"Le parole chiave più importanti e pertinenti alla pagina sono: {top_5_queries}.\n"
        f"Crea un nuovo tag title in {language} che incorpori queste parole chiave in modo naturale e coinvolgente. Il tag title dovrebbe:\n"
        f"1) avere al massimo 65 caratteri, spazi inclusi. Assicurati che la tua risposta non superi questo limite.\n"
        f"2) non includere il nome del brand.\n"
        f"3) non utilizzare lettere maiuscole inutili, simboli, apici o virgolette.\n"
        f"4) essere presentato da solo, senza alcuna spiegazione o contenuto aggiuntivo.\n"
        f"Per favore, segui scrupolosamente tutte queste regole nella tua risposta. Se non sei sicuro, conferma che hai seguito tutte le regole nel tuo messaggio."

        # Modifica la struttura dell'input per utilizzare il formato chat
        #response = openai.ChatCompletion.create(
        #response = client.completions.create(
        response =

                {"role": "system", "content": "Comportati come se fossi un SEO copywriter professionista."},
                {"role": "user", "content": prompt}

         # Aggiungi il numero di token utilizzati per questa risposta al totale
        total_tokens_used += response['usage']['total_tokens'] 
        actual_api_calls += 1  # Incrementa il contatore per le chiamate API effettive

    # eccezioni errori gpt
    except Exception as e:
        print(f"Error occured while calling OpenAI API: {e}")
        return None

    #improved_tag_title = response.choices[0].message.content
    #improved_tag_title = response.choices[0].text.strip().lower()
    improved_tag_title = response.choices[0].message.content.strip().lower()

    #gpt_responses[tag_title] = improved_tag_title
    # modifica per usare come chiave di confronto l'URL
    gpt_responses[page_url] = improved_tag_title

    return improved_tag_title

# chiamate API gpt
# Dopo la definizione della funzione, esegui il loop per migliorare i tag title
merged_df8b.reset_index(drop=True, inplace=True)
processed_rows = 0

for index, row in merged_df8b.iterrows():
    # Passa la riga corrente e il dizionario delle risposte a improve_tag_title
    new_tag_title = improve_tag_title(row, gpt_responses)

    # Assegna il nuovo tag title alla riga corrente[index, "new tag title"] = new_tag_title
    # Aggiorna il dizionario gpt_responses con la risposta ricevuta
    gpt_responses[row["page"]] = new_tag_title
    # Stampa il progresso
    #print(f"\rProcessed row {index+1}/{len(merged_df8b)} | {'%Y-%m-%d %H:%M:%S')} | {total_tokens_used} tokens.", end="", flush=True)
    processed_rows += 1  # Incrementa il contatore di righe processate
    print(f"\rProcessed row {processed_rows}/{len(merged_df8b)} | {'%Y-%m-%d %H:%M:%S')} | {total_tokens_used} tokens.", end="", flush=True)

# Aggiungi qui il nuovo blocco di codice per assicurarti di aggiornare tutte le righe
for index, row in merged_df8b.iterrows():
    page_url = row["page"]
    if page_url in gpt_responses:[index, "new tag title"] = gpt_responses[page_url]

# Stampa il totale dei token utilizzati
print(f"\nTotal tokens used: {total_tokens_used}")
print(f"Actual API calls made: {actual_api_calls}")


Hello, after the recent update my script ended working and I can’t figure out why.

The error is: Error occured while calling OpenAI API: ‘ChatCompletion’ object is not subscriptable

If I may leave a comment, I find it very unprofessional to release an upgrade that breaks the work done by developers until the day before. You are wasting people’s time and nerves.

You can convert the response output object to a dictionary which can be parsed similarly to before:

response_dict = response.model_dump(exclude_unset=True)

Going to close this thread out since it is getting quite long, see the solution for details on fixing this problem.