OpenAI API Unrecognized Messages Argument

Problem

When using OpenAI's gpt-3.5-turbo or similar chat models, developers often encounter the error:
InvalidRequestError: Unrecognized request argument supplied: messages

This occurs when:

1. Using the legacy openai.Completion.create() method with chat models
2. Attempting to pass messages parameter to the completions endpoint
3. Using an incompatible SDK version for chat-based models

The error happens because chat models (GPT-3.5-turbo, GPT-4) require the Chat Completions API endpoint (/v1/chat/completions), not the legacy Completions endpoint (/v1/completions).

Solution

Always match your API method to the model type:

For Chat Models (gpt-3.5-turbo, gpt-4)

Use the Chat Completions API methods:

Option 1: OpenAI SDK ≥ v1.0.0 (Recommended)

Update to the latest SDK and use the new client structure:

pip install --upgrade openai

from openai import OpenAI
client = OpenAI(api_key="your-api-key")

responses = []
for prompt in prompts:
    completion = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": prompt}],
        max_tokens=20,
        temperature=0
    )
    responses.append(completion.choices[0].message.content)

Option 2: OpenAI SDK < v1.0.0

Use the ChatCompletion class:

import openai
openai.api_key = "your-api-key"

responses = []
for prompt in prompts:
    completion = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": prompt}],
        max_tokens=20,
        temperature=0
    )
    responses.append(completion.choices[0].message.content)

Critical Differences

• Use openai.ChatCompletion.create() instead of openai.Completion.create()
• Remove legacy parameters like restart_sequence, top_p, frequency_penalty, etc.

Key Differences Between Endpoints

Feature	/v1/completions	/v1/chat/completions
Compatible Models	gpt-3.5-turbo-instruct, davinci-002	gpt-3.5-turbo, gpt-4
Required Parameter	prompt (string or list)	messages (list of dicts)
Message Format	Raw text	Role-content pairs
Best For	Basic text completion	Conversation workflows
Response Location	choices[0].text	choices[0].message.content

Common Mistakes to Avoid

1. Mixing endpoint types

   # WRONG
   openai.Completion.create(model="gpt-3.5-turbo", messages=[...])
   
   # CORRECT
   openai.ChatCompletion.create(model="gpt-3.5-turbo", messages=[...])

2. Incorrect response parsing

   # WRONG (completions format)
   response['choices'][0]['text']
   
   # CORRECT (chat format)
   response.choices[0].message.content

3. Outdated SDK version
   Always verify your installed version with:

   pip show openai

4. Incorrect message format
   The messages parameter requires role-content dictionaries:

   # VALID
   messages=[
       {"role": "system", "content": "You are a helpful assistant"},
       {"role": "user", "content": "Hello!"}
   ]

Pro Tip

When migrating from legacy models:

1. Replace prompt with messages dictionary
2. Switch to ChatCompletion methods
3. Handle responses at .choices[0].message.content
4. Remove incompatible parameters (like best_of, logprobs)

Troubleshooting 404 Errors

If you see:

openai.NotFoundError: 404 - This is a chat model and not supported in the v1/completions endpoint

1. Verify your model name matches exactly
2. Confirm SDK version supports chat models
3. Ensure you're using openai.ChatCompletion.create() for older SDKs or client.chat.completions.create() for v1.0.0+

Best Practices

1. Always use the latest SDK:

   pip install --upgrade openai

2. Set explicit roles in messages:

   messages=[
       {"role": "system", "content": "Set assistant behavior"},
       {"role": "user", "content": "Your question here"}
   ]

3. Handle rate limits:

   from tenacity import retry, stop_after_attempt, wait_exponential
   
   @retry(stop=stop_after_attempt(3), wait=wait_exponential(multiplier=1, min=4, max=10))
   def get_chat_response(prompt):
       return client.chat.completions.create(
           model="gpt-3.5-turbo",
           messages=[{"role": "user", "content": prompt}],
           max_tokens=100
       )

4. Stream responses for long interactions:

   stream = client.chat.completions.create(
       model="gpt-4",
       messages=[{"role": "user", "content": "Tell me about quantum computing"}],
       stream=True
   )
   
   for chunk in stream:
       if chunk.choices[0].delta.content:
           print(chunk.choices[0].delta.content, end="")