[Enhancement]: OpenRouter Reasoning Traces missing from Sonnet 3.7

[joshbla]

What features would you like to see added?

It seems Claude Sonnet 3.7 is not displaying reasoning traces from OpenRouter.

More details

I assume this is because there is not a default reasoning token budget. I don't detect any delay before the response so I assume no reasoning is occuring.

Ideally, there would be a reasoning tokens parameter slider for both OpenRouter and Anthropic directly.

Which components are impacted by your request?

UI, Other

Pictures

No response

Code of Conduct

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

[danny-avila]

OpenRouter is already configured to interpret reasoning tokens if provided. They don't yet support it.

Extended Thinking coming very soon to the OpenRouter API. Read more about it here, and stay tuned for API updates!!

https://discord.com/channels/1091220969173028894/1092729520181739581/1343669893991628830

[joshbla]

There's a new endpoint on OpenRouter for thinking specifically. In the announcement they say there is a new cross-model reasoning standard they working on. It might be good to hold off on implementing a slider until that is done. The thinking endpoint works in LibreChat already.

According to their current docs the budget is auto-calculated by OpenRouter by setting budget_tokens to 80% of the max_tokens value. As they say "Right now it is not possible to set a custom budget, but this will be added soon."

A slider for Anthropic as a provider exclusively could be done now, just like for the current OpenAI o-model slider (although you have it active for OpenRouter as well), but the question is how many UI elements you want to have (especially if OpenRouter is building a cross-platform reasoning standard as well). Personally, I would prefer more fine-control than less in the end.