Thinking Token Cost Isolation
First-in-world tracking and isolation of reasoning/thinking tokens to attribute exact processing costs.
Reasoning models like OpenAI o1, o3-mini, and DeepSeek R1 generate internal "thinking" or reasoning steps. These steps consume output tokens that are charged at output rates, yet they are excluded from the final visible response.
The Challenge
Since thinking tokens are not returned in the message content payload, standard APM logs miss them, leading to a 30-50% cost misattribution. RaksHex solves this by parsing the usage metadata object in the API response payload.
Attribution Analytics
The Cost Dashboard highlights the exact percentage of your model spend going towards reasoning steps vs standard output, helping developers evaluate if a model's logical depth is worth the extra cost for a given task.