Reasoning Effort set to None still causes excessive reasoning in LibreChat

[AtlantisDe]

What features would you like to see added?

Issue: Reasoning Effort set to None still causes excessive reasoning in LibreChat

I have configured Reasoning Effort to None in LibreChat, expecting the model to reduce or disable reasoning token generation and provide faster responses.

However, during actual conversations, the model still appears to perform extensive reasoning before responding. The response latency remains high, and a large number of reasoning tokens are still being consumed. The behavior is similar to having a high reasoning effort level enabled.

This results in:

• Increased time-to-first-token (TTFT);
• Slower responses in normal chat scenarios;
• Higher token consumption;
• Poor overall user experience.

According to the documentation:

Reasoning models only: constrains effort on reasoning. Reducing reasoning effort can result in faster responses and fewer tokens used on reasoning in a response.
"Minimal" produces very few reasoning tokens for fastest time-to-first-token, especially well-suited for coding and instruction following.

Based on this description, lowering the reasoning effort should limit reasoning behavior and reduce reasoning token usage. However, setting Reasoning Effort: None in LibreChat does not appear to have any noticeable effect.

I would like to clarify the following:

1. Is LibreChat correctly passing the reasoning_effort parameter to the API?
2. Does None mean the parameter is omitted rather than disabling reasoning?
3. For reasoning models, should users select Minimal instead of None to effectively reduce reasoning effort?
4. Could there be an issue with parameter mapping or compatibility between LibreChat and the model API?

Expected behavior:

• Normal conversations should avoid unnecessary reasoning overhead;
• Time-to-first-token should be reduced;
• Reasoning token usage should decrease;
• Higher reasoning effort should only be used for complex tasks such as code analysis or advanced reasoning.

More details

Which components are impacted by your request?

No response

Pictures

No response

Code of Conduct

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

[danny-avila]

This is dependent on the provider actually making use of this parameter, would need more details on what model/provider you are using