Will Anthropic include reasoning tokens via the API? Exposing reasoning in many ways enhances transparency and helps confirm the overall output, but depending on how good it is, there may be competitive or safety reasons to not expose it. But I hope they do!