It would be awesome if the usage statistics in the chat completion object included image token usage, like it does audio. We process a ton of images with gpt4o, and it would be great to receive confirmation from OpenAI on how much of our general token consumption comes from images.
Right now, we need to implement our own image tokenizer (since TikToken doesn’t natively support it) and fire off a separate request to log the “estimated” image token consumption. This kinda clogs our workflow and is much lower fidelity than if OpenAI’s API response confirmed what we actually used.
Thanks! If anyone else has tips on some other way to do this, much appreciated.