A: Recently added our own model much cheaper. The thing is, it's not as simple as just replacing the current model and plugging in another one. Each model has its own unique issues and edge cases that need to be properly handled. Ensuring that every model we use is up to date and doesn't cause any issues is a significant task for us.

For example, we added Haiku recently, but after months of testing, we couldn't get it to stop making things up or expressing its internal thoughts. We decided to remove it because users would choose that cheaper model and then come to us complaining when these issues occurred.

I think for now we have found a good middle ground with our own model. It is much cheaper and a lot faster, yet it still performs on the same level as the pro claude model. We are going to see if we can improve upon this and make it even more cost-effective.