To get full access to the Kademi Hub existing customers can login here, or new customers can register here.
Breakdown of the completion token usage OpenAI returns for a chat completion, distinguishing tokens spent on reasoning, audio and predicted-output handling from the total completion tokens. Nested inside CompletionsUsage as the completion_tokens_details field of an OpenAI chat completion response.
Properties
| Property | Returns | Description |
|---|---|---|
| acceptedPredictionTokens | Integer | Number of tokens from a supplied predicted output that appeared in the model's completion and so were accepted, reducing the effective cost of the request. |
| audioTokens | Integer | Number of completion tokens generated as audio output. |
| reasoningTokens | Integer | Number of completion tokens the model spent on internal reasoning before producing its visible output. |
| rejectedPredictionTokens | Integer | Number of tokens from a supplied predicted output that did not appear in the model's completion and were rejected, and so were still billed as completion tokens. |
Methods
getReasoningTokens()
Returns: Integer
Number of completion tokens the model spent on internal reasoning before producing its visible output.
getAudioTokens()
Returns: Integer
Number of completion tokens generated as audio output.
getAcceptedPredictionTokens()
Returns: Integer
Number of tokens from a supplied predicted output that appeared in the model's completion and so were accepted, reducing the effective cost of the request.
getRejectedPredictionTokens()
Returns: Integer
Number of tokens from a supplied predicted output that did not appear in the model's completion and were rejected, and so were still billed as completion tokens.
© 2025 Kademi Ltd