Right now ResultMessage.usage lumps thinking tokens into output_tokens, but thinking tokens are actually billed differently from regular output. This makes it impossible to accurately track costs.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results