Uncertainty estimates based on explicit confidence statements the model generates as part of its reasoning output.