How to set reasoning level to model in kilocode VS extension v7.1.18 ?

[medphisiker]

Question

Hello,

when configuring a provider and its associated LLMs in the current version of the VS Code Kilocode extension (v7.1.18), it is possible to specify a list of models.

For each model, we can define a Model ID and a display name.

However, how does one specify the "reasoning level" for a model?

Previously—within the profile settings—when selecting a model, it was also possible to choose its reasoning level from a range of options: None, Low, Medium, High, or Very High.

It would be great to bring back this functionality, just as it was before.

P.S. Thank you for your work and kilokode extention updation.

[markijbema]

In the model selector in the prompt box it's right there right? where are you trying to change it? can you add a screenshot?

[medphisiker]

I use specific OpenAI compatipable provider for GPT-5.4 model.

I add and configure my provider.

I get list of model from my provider in the right way.

And I haven't the option to select reasoning level for gpt-5 models from my openai compatipable provider.

[medphisiker]

When I use kilo gateway I also haven't reasoning level for a Qwen3.6 free model.

[medphisiker]

But if I select OpenAI GPT-5.4-mini from Kilo gateway I see the reasoning level selector )

[chrarnoldus]

When I use kilo gateway I also haven't reasoning level for a Qwen3.6 free model.

Adding an on/off toggle now, should be live in an hour or so

[medphisiker]

Adding an on/off toggle now, should be live in an hour or so

Thank you, it works for Qwen3.6 free from Kilo gateway for now.

[medphisiker]

It would be great to have the ability to add reasoning levels for models from OpenAI-compatible API providers—for example, manually via the UI.

[fdterr]

This needs to be available for the openrouter version of many models as well

[medphisiker]

It might be convenient to add an option in the UI for users to specify the reasoning levels supported by their model.

The user can read the documentation and add these values for the model.

This would be very useful for OpenA Compatipable Provider, as the user often knows which models they have and which reasoning levels they support.

It seems difficult to implement automatic reasoning levels for unknown models from the OpenAI Compatipable Provider.