The current implementation of parameter customization is highly restrictive and prevents users from fully harnessing the capabilities of advanced models.

A prime example is the new O3 PRO model. It features a critical reasoning_effort parameter, which significantly enhances inference quality when set to high. However, due to the current system's limitations, users have no way to enable this setting. This means we are unable to use the model at its full potential, which is a major bottleneck.

The ideal solution would be to grant users the ability to fully customize the API request format itself. Instead of being locked into a rigid key-value system, users should be able to directly edit the JSON request body.

This would immediately unlock compatibility with all current and future models, regardless of their unique parameter structures (like O3's reasoning_effort). From a development perspective, this approach is far more efficient. It would eliminate the need for developers to manually add support for every new parameter for every new model. This frees up development resources and ensures users don't have to wait for patches to use new features.

By empowering users with true customization, you create a more flexible, powerful, and future-proof platform. It’s a win-win for both the user community and the development team.