[Enhancement]: Enable Default "Formatting re-enabled" Prompt for ChatGPT O1 and O3 Models

[xmokay]

What features would you like to see added?

Add "Formatting re-enabled" Prompt for ChatGPT O1 and O3 Models

More details

Currently, the default markdown formatting for the ChatGPT O1 and O3 models in LibreChat is quite messy, which may lead to a confusing user experience. Although users can customize this behavior using presets or Model Specs, it isn’t very friendly or convenient for everyone.

I propose that we include the prompt "Formatting re-enabled" as a default for these models. This change would automatically apply cleaner markdown formatting, improving readability and reducing the need for users to tweak settings manually.

Thank you for considering this enhancement!

Which components are impacted by your request?

No response

Pictures

Code of Conduct

• I agree to follow this project's Code of Conduct

[xmokay]

The OpenAI forum has also been mentioned:
https://community.openai.com/t/o1-api-doesnt-separate-code-blocks-with-three-backticks/1099469
https://community.openai.com/t/o3-mini-does-not-format-text-in-api/1113222
https://community.openai.com/t/o3-mini-api-markdown-formatting-issue-with-php-code/1113307/3

Bad case example:

Write code in pyton and compute the first 20 prime numbers

response

Below is a Python script that computes and prints the first 20 prime numbers:

------------------------------------------------------------
#!/usr/bin/env python3
import math

def is_prime(n):
    if n < 2:
        return False
    # Only test up to the square root of n.
    for i in range(2, int(math.sqrt(n)) + 1):
        if n % i == 0:
            return False
    return True

def first_n_primes(n):
    primes = []
    current = 2  # the first prime number
    while len(primes) < n:
        if is_prime(current):
            primes.append(current)
        current += 1
    return primes

if __name__ == "__main__":
    first_20 = first_n_primes(20)
    print("The first 20 prime numbers are:")
    print(first_20)
------------------------------------------------------------

Explanation:
1. The function is_prime(n) returns True if n is prime, and False otherwise.
2. The function first_n_primes(n) iterates over the integers starting from 2, checking for primality and adding prime numbers into a list until it collects n primes.
3. Finally, the script prints out the first 20 prime numbers.

When you run this script, it outputs:

The first 20 prime numbers are:
[2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71]

screenshot

[thefunny1983]

I wonder: Is it possible to use addParams for this? Like mentioned here:
https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/custom_endpoint

[alexyangjie]

In my agent setting, even if I include the line "Formatting re-enabled" in the first line of the instruction, the model is still not returning proper markdown text. I think it might be because we enabled tools and other features that the default system prompt got overwritten.

[Ases]

I’d love to hear the opinion of @danny‑avila on this. With the o3 model coming in about 80 % cheaper, it looks like a good option for wider adoption. The only sticking point is that less‑technical LibreChat users have to remember to re‑enable “Formatting,” which can be a bit confusing.

Is there a way to set “Formatting” to ON by default for everyone in a LibreChat instance, so they get a smoother experience right out of the gate?

[alexyangjie]

Just add a bit of information in case someone wants to pick it up and fix it. @danny-avila

LibreChat/api/server/controllers/agents/client.js

Lines 749 to 755 in 9c70d1d

	let systemContent = [
	systemMessage,
	agent.instructions ?? '',
	i !== 0 ? (agent.additional_instructions ?? '') : '',
	]
	.join('\n')
	.trim();

Here, the agent instructions is appended to the system message, so even if the user wants to enable formatting by adding the "Formatting re-enabled" line in the agent instructions, it is not applied since it is not the first line.

The simple fix would be to move the agent.instructions before the systemMessage. If moving the sequence would introduce unexpected behaviour (since we know that changing sequence of prompt sentences have an impact on the model behaviour), we can have a logic that searches agent.instructions for the specific word of "Formatting re-enabled", and directly add this sentence to the first element of the array.

[xmokay]

GPT-5 has the same issue.